Moderna, Inc. (MRNA) Earnings Call Transcript & Summary

Good morning, good afternoon. Welcome to Moderna's headquarter. Thank you so much for taking the time today to be with us in person or to be with us online. So welcome to Moderna 2025 Analyst Day. As you know, our mission is really the North Star of this company. We want to work together with all our stakeholders to deliver the greatest possible impact to people through mRNA medicine. If you think about the near term, our strategy is really set up on two axes. One is to build a large seasonal vaccine franchise for high-risk population, and two, is the cash generated by that franchise to invest in oncology and in rare disease therapeutic. If you look about the seasonal vaccine franchise, we are already quite underway. We now have three approved products. We have positive Phase III in flu, positive Phase III in flu plus COVID, and Norovirus is currently enrolling in Phase III. And if you look at oncology, 2026 is going to be quite an exciting year with a lot of data in oncology and a lot of new medicine getting into the clinic. PA is now in Phase III, fully enrolled and MMA is Phase III ready. If you think about the seasonal vaccine franchise, we think there's a lot of positive of that business. One, of course, is the tailwind of a growing older population. If you look at the numbers, there are 250 million people right now in the OECD countries that are qualified as older population, 65 and above, and of course, growing with an aging population. And if you look at Europe, and Stephen will talk a lot about Europe in a minute, there are actually 90 million people right now in Europe that are 65 and above. It's a larger population set of countries. And so it's an older continent. So we think it's quite an interesting opportunity for us in terms of tailwind. If you look at the burden of disease, which is why we do what we do, we really think that those products, those medicines are really, really important to prevent disease, but even more importantly, to prevent hospitalization and to prevent death. If you look at the numbers, these are just U.S. numbers. They are very, very large numbers. It's up to 1 million people hospitalized every year. Think about the impact on those people, the impact on those families. Many of us being young do not always appreciate how traumatic a hospitalization can be in terms of how it impacts the quality of your life when you are older, when you have comorbidity risk. Sometimes in hospitalization, even if you survive, it might mean a drastic change of your quality of life because of the muscle mass that you lose during the hospitalization. And if you look at COVID, the numbers are also really high in terms of what we can do. We didn't even spend time on RSV and so on. So if you look at the burden of disease, it is really high. If you look at this franchise, there's some interesting seasonality, which I think is going to be really important for us in the flu plus COVID combo product. We also think it's important with the mRNA technology speed, the ability to potentially go very late in strain selection. And we see from time to time, as we spoke in the past, that you have a mismatch in flu, for example, between the strain pickup by WHO that are in the vaccines and what is circulating around the world, leading to more hospitalization, more disease, and more deaths. And so we think with the mRNA technology, we have shown now several years in a row that we can be informed by the FDA of the regulators in May or June, a new strain, and we can very quickly get ready in front of a season, which I think is a huge advantage of the mRNA technology. And then there's a lot of other benefits. One of our platform is, of course, manufacturing scale-up, and Jehr in a minute will give you a lot of details around that is because mRNA is a very flexible manufacturing process. We can literally a week make a COVID lot and then a week after make a flu product. So as you see over the next few years, as Stephen will describe the product launches, as we get more and more volume, we're going to get an incredible leverage of this manufacturing infrastructure that we have. The market access and reimbursement are pretty set up because healthcare providers know the burden of disease. If you talk to health ministers, they know that with an aging population and respiratory viruses, they have a massive financial challenge ahead of them. The life cycle investment is very manageable because those products have a very long tail. Of course, you need to update them. Sometimes you need to do seasonal studies, but this is very manageable. And in the U.S., we have a very interesting setup where there's a strong incentive from the channel, especially retail pharmacies to drive vaccination every season. Stephen will discuss in detail why we're excited about the next few years. What you're going to see, we set up a portfolio driving diversification of growth through new product launches, like you're going to see mNEXSPIKE with a full year next year, launches full year in U.S., launches in the rest of the world, potentially flu launches in '27 and then Norovirus flu plus COVID, and then also geographic diversification. I know sometimes people are worried about the U.S. But if you think about it, we have now a great partnership in U.K., Canada, Australia with full year impact next year. And then the European contract reopening for us the ability to participate in Europe, which is a very large market in '27. So we're going to see product growth through launches and also geographic diversification that's going to drive growth in the next few years. Jamey will walk you through some numbers. And if you look at that respiratory franchise, you're going to see the next few years growth. The margins -- gross margins are also going to improve because you're going to have additional volume in an existing infrastructure. And as Jehr will share with you, the team has some great productivity improvement, and there's more to come. There's a lot of exciting projects that Jehr and his team are leading, which will drive improvement in gross margin with those two factors. And then as we've been talking for a year or 2 now, the cost of R&D for respiratory disease are going to go down. And they're going to go down because those very large onetime Phase IIIs that you have to invest to build that franchise are basically concluding. They don't conclude the day of launch because you have a tail of investments as you have to finalize those studies, monitor the safety, which is very important. But when those are over, you have basically a onetime investment, which is very large, which is a great barrier to entry. And then you have many, many years, if not decades, where you can have those products driving revenues. So when you look at this plus commercial in the U.S., it's mostly a B2B setup where we have direct access to the retailers because we don't go through PBMs in vaccines. We do not need to add teams as we basically grow the portfolio. And actually, as we spoke in the past, a larger portfolio is a strength to our ability to get great contracts and talk about the products with the customers. So if you think about it, growth on the top line, improvement in gross margin, lower R&D cost and very stable through existing infrastructure, SG&A. It's very interesting improvement in operating margin that we're going to see over the next few years. If I now turn to the future, where we're investing the capital that we are generating through the vaccine franchise into oncology and into rare disease. And again, the team will spend a lot of time on the details, so I want to be brief to start, 30,000 feet. You're going to see potentially Intismeran with a launch in '27. As we talk about, the first IO readout is in '26. And if it's positive, we are working very closely and diligently already with the Merck's colleague to be able to file very quickly, which could lead to a launch in '27. If you think about PA in rare disease, PA is now fully enrolled for his registration study. And so we could see PA launch in '28. And then for those of you that have not followed at ESMO or were not at the Analyst Day at ESMO, we shared some early but quite exciting data on 4359 in Stage IV melanoma and lung patients. The team will walk you through some data. Because of that data, we accelerated the Phase II this year to be able to see is that signal real? Can we confirm it with the larger numbers because that could be a very important product for patients and the product that is 100% owned by Moderna product. So that's really for product that are in late stage. If you look at earlier-stage product, as I said, MMA is ready to go into Phase III. We have a couple of more exciting products in oncology, and the team will walk you through it. We have an EBV treatment program that could be very important for patients with multiple sclerosis. An EBV vaccine that's also with interesting data that could be used potentially for mononucleosis infection, Lyme disease as our first bacterial vaccine, and so CMV for transplant. So those are the two franchise we're really focused on, generating cash through a lot of financial discipline, sales growth, respiratory vaccine, investing the cash for innovation for the next wave of products. So we could see sustained sales growth for many, many years to come, thanks to our platform. As you've seen, what we've done in the last couple of years is a lot of financial discipline. And I'm so thankful for the teams for what they have done under Jamey's leadership and all my colleagues at the EC to grab cost across the entire enterprise. And the momentum that we have is very exciting, as you saw on the Q3 earnings call, which gives us confidence for the ongoing work and projects to come to be able to get the cost into the right place to drive profitability in 2028. Jamey will walk you through it. There is a lot of productivity projects. There is a lot of digital investment, and there is a lot of AI tools also being worked through. For those of you who are going to stay after lunch, we have set up a bunch of panels to walk you through just vignette of examples where you have members of the team will come here to walk you through some of the tools we are using every day in the business to drive productivity through AI across the entire enterprise. So with this, I want to now turn to Stephen to walk you through the more detail of the seasonal vaccine franchise. Then Jehr is going to walk you through manufacturing before Jamey wraps up everything again at the enterprise level around financials for the next few years. And then the team will come. We start with seasonal vaccines with Jackie and her team. Then we'll go into early vaccines before we take a break. And after we're going to have a lot of exciting things to share in oncology and then rare disease before I have a very quick two-slide close, and then we'll be very happy with the team to take your questions. So with this, I will turn it back to Stephan.

Thank you, Stéphane, and welcome, everyone. Hello. It's good to see you again. I want to quickly summarize Stéphane just mentioned. There is a diverse portfolio of drivers of growth that we think will contribute to that steady financial performance we're looking for over the next few years as we invest in accelerating our company. Stéphane already presented this slide. I will just highlight that we do have a mix of geographic and product growth drivers. And really in the next 2 years, it is mostly geographic or market expansion into U.K., Canada, Australia, into Europe and into new regions, including Latin America. It's really as you look to the second and third years of the next 3 years that we'll expect to see a substantial uplift from investments in our portfolio. And that includes flu, our flu/COVID combination product and Norovirus. And I'd like to click through those and just give you a sense of why we are bullish on these opportunities. So starting in '26. We have multiyear strategic relationships with the United Kingdom, Canada, and Australia that we have been working on for the last few years. And so announced most of them in '23. And this last year, in 2025, we have licensed the three facilities that will actually supply those agreements. As a reminder, these are long-term multiyear contracts. They involve us investing in domestic research and development, and they are part of a national security and biodefense as well as public health strategy at the core of each of these countries' approaches to protecting against viruses, but also protecting against future pandemic threats. That onshore manufacturing is now licensed and starting to deliver products. In the U.K., that covers 69 million lives. And we had already referenced in prior financial statements that we will see that first approximately $0.2 billion of revenue from that partnership in the first quarter. That was delayed from the fourth quarter of this year into the first quarter of next year, but we'll also be delivering vaccines for their fall campaign, first with COVID, but then eventually with the rest of our respiratory portfolio. Canada, again, 41 million lives. We do have strong performance this year in Canada are already delivering made in Canada COVID vaccines, hope to add additional products, RSV and mNEXSPIKE to those relationships. And we expect to see a full annualized impact from that agreement starting in 2026. And then third, Australia, 27 million lives, where, again, we've now licensed that facility. We're hoping to make our first deliveries possibly even this calendar year. And as we look to 2026, see the annual benefit from the strategic partnership. So all of this in place and a major driver of growth for us as we look to 2026. The other driver for us that I'd highlight is mNEXSPIKE. MNEXSPIKE was our successful launch this year. I will remind you, we've only really launched this product commercially in the United States. And the product was only approved mid year, really in June, and we had a tremendous amount of work through the summer to get ready for the fall season, not just in manufacturing, but in market access and reimbursement and preparing the market for it. And so we're incredibly excited to see the market share that we've already grabbed in this first year of launch. 23% of the total retail shots in arms year-to-date has been from mNEXSPIKE. And actually, if you look in older adults, those 65 and above, that market share is almost 1/3, a really strong start. It has become our leading product in the retail channel as well as in that 65-plus and high-risk market that we really think is the future for us in COVID. Now mNEXSPIKE, as we look to 2026, we hope to complete that launch. We've only really had a half year this year, and we look to drive even better performance in the U.S. And we expect to add many other markets. Europe, we're under review and hope to be approved. We've recently been approved in Canada, although we haven't launched the product yet this year commercially. And Australia, Japan, and Taiwan, all have submissions pending for approval. So we are looking forward to continued strong momentum. Behind this, we believe differentiated profile for mNEXSPIKE going into 2026. Okay. So those are the two drivers for '26. As we look to '27, Stéphane already highlighted the role of continued market entry for us. So first, in the European Union, we have been excluded from that market largely because of a pandemic era contract that is now falling away in 2026. If you look at the size of that market in the most recent year for which we have data, so '24, there's about $1.8 billion of respiratory vaccine sales, $700 million from COVID, about $1 billion in flu and an RSV market that's just starting to grow. This '24 was that first year of launch. We do expect that RSV market will get bigger as it expands. We have had a very small share there because of that exclusion from the market, only about less than $100 million. And so clearly, an opportunity for us as those barriers fall away. So why are we excited about '27? Obviously, the end -- the lapse of that COVID contract gives us an opportunity to compete for share in the COVID market. We will also have, we think, one of the most diverse portfolios. By '27, that means an RSV vaccine that's approved. We hope also mNEXSPIKE, a combination flu COVID and flu. In fact, five different products. And we hope in that combination product, the most differentiated an opportunity to compete in this very large, very established market where there's a high burden of disease, particularly in older and high-risk populations. So we're pretty excited about that as a market entry opportunity. We're not done there. While Europe may be the largest market we'll be entering, we are also looking to add additional strategic partnerships. And I'll highlight one here that we announced -- the government of Brazil announced, September 5, just 2 months ago, which is a productive development partnership, which has us transferring and manufacturing with a partner in Brazil, Instituto Butantan for the delivery of COVID vaccines. We hope that starts delivering products as soon as 2027. It's the kind of multiyear strategic relationship, which allows us to get access to that market, but also grow our share and revenue opportunities. And so that was a recent announcement that we're excited about. And there are several others that we're engaged on across Latin America and Asia Pacific right now that would provide strategic access and growth of access for our products over multiyear agreements, much like we're doing in U.K., Canada, and Australia for '26. The other thing we'd highlight in '27 is we'll continue to add product portfolios. And so flu vaccine, we have filed for -- we are in the process of filing for approval across the U.S., European Union, Canada, and Australia in -- by January of next year. So just in the next 2 months and some of those filings implied as we speak. Flu is a very large and established market, approximately $6 billion of sales, growing double digits. And there is a large enhanced vaccine market, $2.5 billion in the most recent year. We think that will grow to almost $3 billion by 2032 that we believe 1010 is well-positioned to compete for. And so again, a place that is, for us, a new opportunity. It is competitive. We recognize that. But through our strategic partnerships and the other infrastructure we're building, we believe we can go compete for it. And anything we get out of the flu market is growth for us. And by '27, we do hope that becomes a substantial contributor. Okay. So those three drivers in '27, mostly market entry and perhaps the beginning of an impact from flu. And then as you look to '28 is when we really expect to have the biggest impact from our pipeline investments. That's not to say that we don't hope these products aren't approved earlier. We do, and we will work to launch them as soon as we can get them approved. But in most cases, it is how do we prepare to really drive growth, and we think that is more in the 2028 time horizon. So first among those I would highlight is the flu-COVID combination product. We've had very exciting data we've shared previously. I know the team will walk through some of that now. We believe flu-COVID combination remains a substantial growth opportunity for us, both in the flu space and the COVID space. And some of the data, the reason why it's highlighted here. If you look in the most recent season, 2024, 2025 season in the U.S., where we have the best data on this. I'll draw your attention to the lower right-hand corner. 47% of those over the age of 65 who got a flu vaccine in that channel also got a COVID vaccine in that same day. It probably won't surprise you to know that when we do market research, people would prefer one shot over two, one arm over two. And we see that as a real opportunity, both to increase the coverage. If that 47% could be higher, that means there's 53% of people that only got one shot and some of them because they only want one at a time. There's an opportunity to expand coverage for COVID and grow that market. But there's also an opportunity to provide the convenience and cost savings associated with just that single injection administration. So we really do see this as a growth opportunity for us across our seasonal vaccine franchise. We have filed for submission in Europe, and we are on path with that review with the European Medicines Agency. So the first potential approvals could actually be coming in '26. EMA was filed last year -- or sorry, in '25. Health Canada has been submitted for approval, again, supported by the new flu efficacy data as a part of that just recently. And so again, in '26, those might be the first markets that we see approval. We're highlighting this as a '28 growth driver because we'll have work to do on market access and reimbursement and preparing to launch those products. But actually, we think we're in good direction there in those markets. And then in the U.S., where we do expect to engage with -- we're engaging with the FDA, we're waiting further guidance from them on what they would like to see prior to refiling the 1083 program. But again, by '28, we do hope to have gotten that approved and be driving growth across our regions and markets. And then the last I'd highlight is Norovirus, which we'll talk a little bit more later today, I'm sure. Norovirus, we see as a very big opportunity. There's a high unmet need, currently no products, over 155 million people in this country who either because of risk factors or age or occupational health risks or some lifestyle choices, they will want to be protected against Norovirus. It is an opportunity for us to add another product into an established channel. We do believe most of the vaccination will be through the retail channel, and it continues to be a seasonal disease. So it really fits into this portfolio that we've been building. So as we look forward to those growth drivers, we do think we are incredibly well-positioned, both from a market entry perspective and a product growth perspective. Focusing just first on the U.S. We have established the relationships with all of the key customers we need, our retail pharmacies, our government customers as well as doctors' offices, hospitals and integrated delivery networks. The strength of the mNEXSPIKE launch this year for us really highlights how we've really matured those capabilities and a good position. And we don't believe we need substantially more capabilities as we add products in the United States market. U.K., Canada, and Australia with those strategic partnerships are well-positioned to deliver growth in '26. And as we add additional products, we will be adding value, but not a huge amount more cost. We don't need dramatically more capability. We'll add value to those strategic relationships. But all of those products can be made in the same facility. And as you've seen and you've heard us talk about, I'm sure Jehr will talk about in a moment, our technology gives us incredible leverage and opportunity to do that. One place where we'll do some investment in growth over the next couple of years is in the European Union to expand that capacity because, again, it is a very large, very established market, we have some capabilities. As I said, we've been generating some revenue, but it's a place that we do expect to see growth, and you'll see us invest in commercial in a disciplined way as we try to grow our presence in the European market, both with the current products and with new products as we move forward. So we think we're well-positioned on the current infrastructure to deliver this plan. There are places where we will have to expand as we go and make additional investments. Stéphane has referenced these, but just to quickly say, we are incredibly excited by the momentum behind Intismeran. I know the team will come and present that data. We do hope for first potentially pivotal data next year in 2026. And then we must get ready, if that is positive, to be launching the product in '27. Those will be some additional capabilities. But we have the benefit of Intismeran from a commercial perspective of having a partner in Merck. And so we work together with them to prepare that for 2027. As you look to 2028, we do expect to be launching our first rare disease program, propionic acidemia. That pivotal study or the pivotal Phase II portion of that study is fully enrolled, and we're looking forward to that data in '26, as you'll hear about. And by '28, we do hope that is a driver of growth for us as well. And then our wholly owned 4359 program, where, again, Stéphane referenced the Phase II data, and I'm sure the team will give you a sense of that preview of why we're enthusiastic about it, but something we would need to be ready for in '28. So while we're really focused on that respiratory and seasonal vaccine franchise, and we do really feel strongly that, that sales growth drives us to breakeven, we also recognize there are opportunities, perhaps requirements that we invest a little bit above that to drive growth in some of these other areas, particularly oncology and rare diseases. And then Stéphane referenced the rest of the pipeline. But I would just say, as you look to '29 and beyond, there are even more growth drivers, even more important product contributions that we expect and where we do expect we will get some leverage from a commercial and medical function perspective. In rare diseases, we'll add MMA. It is a very similar disease and commercial footprint to propionic acidemia. So we believe as we launch that product, we'll be able to use the existing infrastructure. We have multiple oncology products, oncology therapeutics in early-stage disease, and we'll look to be very efficient as we build out our oncology presence. But Rose and others will walk through those programs and why we're excited about them. But as we get ready for '29 and beyond, we do believe we'll have the infrastructure to launch them. And then the early-stage vaccines, these are programs where we are paused in Phase II often, waiting for that breakeven moment so that we can accelerate some of these programs, EBV, Lyme, and CMV for transplant. We'll share some of that data today, some of why we're enthusiastic about those. All of those, we believe, will fit really comfortably into our seasonal vaccine franchise and commercial capabilities that we expect to build over the coming years. So with that, I'm going to turn it over to my colleague, Jehr, who will walk you through how we think some of that technical operations really enables that delivery and scale.

Thank you, Stephen. Good morning, everybody. Welcome. It's really great to have you here and everybody online wishing you a very good day. Our technology gives us incredible leverage. That's the word that Stephen said. And so I want to take you through why and what is about our technology that's really important. And what is it that allows us to drive great efficiency. I think efficiency when I look at manufacturing, it is tremendously important, not only when I think of efficiency that we can deliver speed, speed to market, but that we also have the capability of ensuring highest possible quality and ensuring the highest possible cost efficiency. And that's what we look at when we look at our manufacturing network. So let me take you through a little bit what the network now looks like. It wasn't that long ago, actually 2023 when we had over 8 CMOs globally between Asia, between Europe and the United States manufacturing for us. And part of that manufacturing process was they would manufacture to a demand. And as the demand would change, there would be a take or pay. So it started to get quite expensive. And it was that, that we started to pare down our CMOs to reduce the amount of external CMOs based on quality, CMOs of really high reliable quality, great speed and great cost efficiency. And at the same time, being able to leverage our own manufacturing. So here in Massachusetts, we have Norwood. And in Norwood, is where we make our drug substance, which is the active ingredient, the mRNA for the medicines that Stéphane indicated earlier and Stephen indicated. And maybe anecdotally, let me tell you about speed. It was so exciting to hear Stephen talk about mNEXSPIKE-1283. It was by us having the ability to manufacture that in Norwood, allowed us to have great speed. The ability that already in June, we're identifying what the variant is and then actually have the drug substance made and be able to supply it in the United States and have the channels filled already in 2026 was a huge achievement and not only driven by the fact that we had the manufacturing here in Norwood, but to the amazing people and teams that we have driving that. And that's one of the big benefits that we have now in relation to having drug substance in Norwood and then using ROVI as our contract manufacturer in Spain. And that's the network that's allowed us to drive the cost efficiency we have and the supply we have, not just the United States, but globally. We have now added, and I'll touch on them a little bit later, three manufacturing sites. Stephen talked about these three sites. Between these three sites, we have Laval in Canada, we have Harwell in the United Kingdom, and we have Clayton in Australia. And as Stephen indicated those three sites serve a population of 137 million people. And these are a long-term partnership and agreements that we have with the United Kingdom that allows us a revenue stream that we can rely on and the supply reliability that these countries can rely on from us. And then we just announced and you've read it, a new DP manufacturing in the United States. We are tremendously excited about that for many reasons. If you just look at the logistics of making drug substance in the United States, then sending that drug substance to Europe, fantastic partners we have in ROVI and then sending that back again to United States, operating not only geographically across the Atlantic twice, but also within two different quality management systems. We also have, as I mentioned earlier, when you work with CMOs, we've exited quite a number of them because we wanted to optimize our take or pay. And that's what drove us to announce as we did yesterday, a new DP manufacturing facility here in Norwood. So what this allows us to do is we now can manufacture drug substance in Norwood. And then in the same site, we can manufacture the drug product. So not only taking out all of the logistic time, but you're operating under one quality management system, which means you don't have to wait for a release and approval before it leaves your quality management system and then goes into another because you're in one quality management system, you can move your product through the system much more efficiently. So from a point of view of lean, you can remove a lot of waste time. So not only then does this allow us actually even greater speed to market, even greater ability to optimize our channels, but also with Stephen's team from a commercial point of view to actually fine-tune our manufacturing based on the demands that are there. It's as close as you can get to real-time manufacturing based on demand. We are super, super excited about that. The second reason we're super excited by it is the efficiency. So if you look at drug substance manufacturing, we do a lot of that manufacturing already in the earlier months of the year. And then in the mid-months of the year, you're doing the drug product manufacturing. What we'll be able to do in Norwood is optimize our manpower to be able to work with our drug substance manufacturing operators, be able to train them to do drug product and start moving our people around. Then we have the one quality management system. We have the one quality assurance, the one quality control, the one engineering and facility services. So the synergies we get from that is absolutely huge. And also, we are building out in a facility that already has the infrastructure. So we didn't have to actually break ground to build a new infrastructure because we have optimized our footprint, we were able to build this in an existing footprint already. We had already prepurchased some of the equipment. So from a cash out point of view, a lot of that cash is paid for this facility. So it's very efficient going forward. Our total CapEx over the next number of years is not going to change. This also allows us to drive a lower cost per dose. So it allows us to be in a position as volumes increase based on what Stephen presented to continuously drive a gross margin expansion, which is critical for us. So I'm so excited by this because I think the -- I think honestly, what manufacturing delivers in Moderna is the ability to have top-class speed, the ability of highest possible quality and at the same time, the greatest possible cost efficiency. That's the magic triangle for manufacturing anywhere. And by us taking control of all of this allows us to optimize while at the same time, being able to partner with a CMO so that we have the optimized efficiency for Europe, as Stephen already indicated, without having a large number of CMOs. So we're in a very, very exciting part of the manufacturing journey of Moderna. It wasn't that long ago, not only did we have eight CMOs, but that we were making a lot of our drug substance and Lonza in Europe. That was not long ago. It wasn't that long ago when we had only one manufacturing site, which was Norwood trying to do all of this. Now we got Norwood drug substance and drug product end-to-end here in the United States, made in the United States to supply the United States and supply other countries as well. We're tremendously excited by that. And I see Tracey is here with us. We speak a lot about our physical assets, and we speak a lot about our digital assets and Tracey manages our digital assets. But I don't want to forget our people. We have top-class manufacturing associates that I would class as the best in the world are definitely up there, who are behind all of this because it takes great intellect, great design, and great ambition that we have in our Moderna mindsets to be able to make this happen. So this is one of the things we're super excited by. What's amazing when you look at this slide is not that we have three sites in the United Kingdom, in Canada, or in Australia. I think what's amazing in this is 2 years ago, this was land. It is amazing to be able to be able to build out these sites so fast and then to have them licensed already and to be in a position that they're already supplying, already supplying the 137 million populations of these three countries. This is a huge call out to the engineering team and the design philosophy we had, which was state of the class excellence, the ability to use robotics, automation, to have a learning mindset, so to look at all the experience we had from working with Lonza, look at all of the experience we had from working with CMOs, look at all the experience we had ourselves from working in Norwood and design that in so that these efficiencies are super efficient in relation to manpower and batches per person, but also super efficient in how we can use AI, robotics and automation already from the start and not having to design it in later. And what's also unique about these sites, and I would argue you probably won't find any place else in any other manufacturing that have these three sites are exactly the same. So if you were helicoptered into one of the sites, blindfolded, you wouldn't know which sites you're in, unless you could look out the window and see which flag is flying. They're designed exactly the same way. And there was a huge benefit in this because we started off the installation and operational qualification in Canada, then on to the United Kingdom, and then to Australia. So as we started to do installation qualification, operational qualification and as part of that process, you find issues and you correct those issues, we were already then able to pre-correct them in the United Kingdom and pre-correct them in Australia so that those facilities were even up and running even faster. And then we partner in Lavelle in Canada, and we partner in Harwell in the United Kingdom with CMOs so that we have a full integrated vertical supply for those countries. And in Clayton, Australia, we make our own DP because we weren't able to find that capability there. In relation to operational excellence, we are designing these facilities with optimal cost optimization. So the type of COGS that we're going to deliver out of these facilities will be pretty much exactly the same as what we're delivering here out of Norwood. We have the same one management, one quality management system, one manufacturing system, MS&T, quality assurance, engineering right across all of this so that we can drive efficiencies through this. And so as I speak about efficiencies, let me touch a little bit about margins. I think what you might find impressive on this is that despite a 40% reduction in total revenue across those 2 years, we were able to keep gross margin improvements pretty much flat. And why and how? Firstly, that was driven by removing take or pays, starting to reduce down the amount of CMOs we have. We were able to take a lot of cost out of the system with that. The second thing we were able to do is drive a real ambition in relation to efficiency, robotics, automation. We were starting in Norwood with five operators per suite, and we got -- we set ourselves targets for 3. We got to 2.2 and then hit 1.9. We drove an amazing efficiency. And that type of efficiency that we have in Norwood is what we have now sent out to the other sites. We are then able to apply what I would call lean, but modern lean, the curiosity of understanding the manufacturing process in great detail so that you can manage the quality envelope and that you can ensure that you're right first time. Our drug substance manufacturing, for example, this year in Norwood for all of the products that Stephen indicated, the 1273, the 1283, the mNEXSPIKE and the Spikevax, 100% success rate, no batch out of spec. That is -- and not only do we achieve that this year, but also we achieved it last year. In fact, last year, we had one batch out of spec when we pivoted when the FDA wanted to change the variant, and we stopped that manufacturing of that batch overnight. And the next morning, we were manufacturing as well. So this is what gives me the confidence as when we see -- and Jamey is going to touch on it later, as the volumes start to grow, we'll be able to drive an enhancement in gross margin because we've got the system in place. Also in relation to inventory and inventory management, we've made great progress. What I would say is a lot done. And what Stéphane said earlier in the presentation is a lot more to do. We have great ambitions ahead, and our game is not finished yet. There's a lot more ahead for us. So that's a little bit about that. So the other element of manufacturing is the personalized manufacturing, which is Intismeran autogene. This is where we have the ability to manufacture personalized per patient. And the previous type of technologies that had this was CAR. CAR technologies always had was a challenge on the COGS and the challenge on the manufacturing COGS. So we built out Marlborough. Again, 2 years ago, Marlborough didn't exist. It was basically a brownfield site. And we built out Marlborough, not with just the ability to supply fast and in relation to supply fast, we already have clinical supply of Marlborough in September of this year, 2 years later, but also the ability to ensure that we're managing the COGS right from the start. Because usually, what happens is you focus on getting supply, you focus on regulatory, then you go, we've got that done, let's work on the COGS now. We started working on the COGS before the facility was even built. And I think it's that level of just ambition and determination to really keep efficiency at first and center is what allowed us to do that. So in relation to Marlborough, what we have is we have an end-to-end operation that's focused on cost, supply, and, of course, quality. And that basically is what we're going to look at as we build out Marlborough. So the second thing when I say as we build out Marlborough is we made sure we didn't put all the capacity in there that we need for 5 or 6 years' time. We set it up so that we can do line by line. So we have the ability to run out 7 lines, and we start with one, and we will maximize that line. And then as that line reaches its full capacity, we will then have line 2 ready. Line 2, we will have a higher standard, and I'll touch on what I mean by that later. And then that higher standard, we will then go back and retrofit in line 1. And then the same later for line 3. So not only have we the capability of adding the capacity only as we need it, so we're not carrying a depreciation shadow that's unnecessary, but also that we have the capability to consistently improve our manufacturing capability without having to do a major redesign. And I think that's super, super efficient when you're charting out a journey to COGS reduction. Again, as I said earlier, we were planning and designing cost efficiency right from the start, not something that we wanted to do later. Not only are we really, really proud and again, an amazing team there, but not only are we really, really proud that we're supplying clinical supply out of this facility already in September, but we're very confident that we've got a really great path to a COGS reduction here. So this might give you a little -- it's a sort of a very simple cartoon example, but it's illustrative just to show you what I mean. When we started off the manufacturing of INT in Norwood, we had an equipment footprint that was 120 square feet. So if you look at this cartoon, you see a picture of a person, and that was the size of the equipment roughly to manufacture for a patient. Several pieces of equipment involved. A lot of it once only use and then destroyed. So it had a COGS that was reasonably expensive. We've already in the current configuration in Marlborough, reduced the size of that down by 3, driven efficiency in the design. So we're constantly looking at the design, constantly looking at what can be reused, how we putting clean in validation, how we work with the FDA on that, how we do briefing books with the FDA on that. And so we've already made huge progress. So this is what we have already in our Marlborough facility. And already in our technical development facility in Norwood, we have already the future configuration, and we're already -- not only have we already designed it, we're already running pilot scale batches on it and testing them. Again, a threefold reduction in size. So not only does that mean then that when line 1 gets full, but line 2 as we will build out, line 2 will be based on the future configuration and then we'll retrofit that. So that we have not only the ability to drive increased capability for the demand that's ahead for all of the different -- 8 different INT products that are all the way through the different phases of the clinical trial, but the ability to drive in COGS. And I think that's something that we've learned also from looking at the other industries and what happened in the whole CAR technology area. So we're pretty excited by that. And I'd probably finish up with probably three things. Jamey likes three things. I think the first thing is just a huge call out to our people. We have amazing people behind us. what gives me confidence that we'll be able to continue this journey. I think the second thing is that we have dependability, not only in relation to quality, not only in relation to supply, but also in relation to cost management. And I think lastly is we never finish. We are ambitious. It's never good enough for us. It's a lot done and more to do. That is our attitude. So we have plans ahead that can continue to drive not only the gross margin expansion improvements that I've indicated in the slide, but even better, particularly as we use AI and robotics to drive that in. And you'll see some of our team later -- we'll take you through some of the AI and robotics that we already use. We already have agentic systems up and running in our manufacturing facilities, and it's quite exciting to see. And so let me say my last word in relation, I'm quite excited to see. I'd love to bring you all to Marlborough and maybe that's for another day. But let me give you an indication of what the site looks like in a 1-minute video. [Presentation]

That's real. That's happening right now, and thank you very much for being with us this morning. A real pleasure to hand over to Jamey.

Thanks Jehr. All right. So let me also extend a welcome to everybody. It's great to see many of you, and thanks for joining live or on our webcast. So as Jehr said, I do like to cover three things. The first is I will talk through our -- where we're headed in 2025. I'll provide a recap from our recent earnings call. Then I'll talk through our financial framework for the coming years. So Stéphane laid out the business strategy, Stephen laid out the commercial strategy. Jehr just walked through our manufacturing strategy. So I'll put a financial lens to all that as it pertains to the next 3 to 4 years. Then I'll wrap up with capital allocation. I'll really take that from an R&D perspective, which is where we are driving most of our capital from an investment perspective. I will then take you to a cash perspective and walk our cash balance over the coming years. And then I'll talk about our exciting announcement this morning, which was the loan that we just announced, which makes our balance sheet even stronger. So I'll get back to that in a few moments. So starting with 2025. As a reminder, we narrowed our revenue guidance to $1.6 billion to $2 billion. That's comprised in the U.S. of $1 billion to $1.3 billion and outside the United States, $600 million to $700 million. So inside the United States, we had provided for -- there's only one large variable left, and that is vaccination rates. And we had provided for within this range in the United States that vaccination rates would be down 20% to 40%. And at the time of our call, season to date, we were down about 30%. We're now 2 weeks later, and it's down 28%. So it's a little bit better than at the time of our call. So we feel very confident within the range in the United States as we're in the middle of November now. Outside the United States, it really just comes down to two things: delivery timing, whether the deliveries will happen inside the fourth quarter or shift to next year. And then in a few countries, vaccination rates as well. So I think the punchline here is after another 2 weeks, we still feel very confident in our guide of $1.6 billion to $2 billion of revenue. From a cost perspective, we said we would take our GAAP costs down by another $700 million versus what we had said at the second quarter. So overall, it's over $1 billion out this year. We focus a lot on our cash costs, which you can barely see here, but on the bottom of that slide. We started the year believing that we would invest $5.5 billion. And now we believe our guidance at the midpoint is $4.6 billion. So from a cash cost perspective, Stéphane mentioned it as well, we are taking out nearly $1 billion of cost versus what we thought at outside of the year. We said we'd make improvements to 2026 and 2027, which I'll get into in a few moments. And all of this is really important as we target cash breakeven by 2028. So that's a quick wrap on the year. From a financial framework perspective, I won't go through this chart again. Stephen and Stéphane have covered it. When I step back from it, I see 10 opportunities for growth over the next 3 years. And we will guide future years at a different time. But right now, in 2026, we believe we will start growing, which is exciting. And we believe that growth is up to 10% in the year 2026. Come our 4Q call, we'll give greater specifics around that. We want to see where the 2025 lands, and then we'll give a little bit more specifics around what the actual guidance is for next year, but you can expect that we will grow starting in 2026, and we believe every single year thereafter. So I want to take a minute on cash costs. I said we were going to make improvements. So our previous estimate for 2025 was $5.1 billion for 2026 was $4.7 billion and for 2027, $4.2 billion on a cash cost basis. You can see the GAAP cost numbers at the bottom. Basically, in a year, we're basically advancing 1 year. So with the previous guidance that I just mentioned of $4.6 billion, we're already beating what we had thought we would do in 2026. And then we're going to do that again. So in 2027, we thought it'd be $4.2 billion. We're actually going to achieve that in 2026. And then finally, we're going to actually take 2027 down to $3.5 billion to $3.9 billion. So at the midpoint, that's $3.7 billion, which is another $0.5 billion out. That is due to all the tremendous work across all the teams. Basically, if I step back when we laid out this plan a couple of years ago, we thought we were at $9 billion in cash costs. We're going to hit $4.6 billion in 2 years, and we'll hit $3.7 billion. We knew that we could do this. We didn't know that we would accelerate the timing as much or that it would be as significant as what the teams have been able to achieve. There are loads of examples across all the levers we talk about in terms of manufacturing efficiencies, procurement savings, what we can do to make our trials more efficient as well. And so the teams are just doing an exceptional job, and they're actually beating our own expectations. So therefore, we have confidence in being able to take 2027 down to something that starts with a 3. So that's it for the kind of financial framework over the next few years. I want to move to capital allocation. And our primary driver of investing capital is in research and development, which will evolve over the next few years. And I think you can see two things from how it evolved. Number one is it will go down. That is largely -- we've been saying this for a long time. As we complete the Phase III trials in infectious disease, those will no longer be there come 2027 and 2028. I'll come back to that in a moment. And then I think the second observation is you can see the percentage of investment is going to increase in oncology. So we believe as we scale down and have a portfolio of hopefully 6 products in infectious disease by 2028, that will just require a basic maintenance of life cycle management for infectious disease to maintain that portfolio. And then you can see the large balance will be in oncology. The last couple of things I'd say on infectious disease is you can see it remains a little bit elevated in 2026 and 2027 for two reasons. One is our Phase III Norovirus trial. We're going to initiate another cohort this winter. And so therefore, we can see still -- that will cover -- carry into 2026. And then our post-marketing commitments as it pertains to COVID. So we do have those in the budget. Those will go into 2026, maybe stretch into a little bit of 2027. And then largely, that's it. That's it from an infectious disease standpoint. All the Phase III trials will be completed, and then we'll get to that life cycle management. From an oncology perspective, we are super excited to invest behind this. We're obviously super excited to invest behind infectious disease, but that is completing. So everything that you're going to hear today from Intismeran or mRNA-4359 or our entire emerging oncology portfolio that Rose will walk you through, that will increase over the coming years. And we are still investing behind rare disease and autoimmune, but we want to see what those -- how those registrational studies read out before we continue to invest further. And either way, it's a relatively smaller investment versus from a patient population perspective as well as dollars perspective versus what we'll see from oncology and infectious disease. So that's how our R&D is evolving, and that's where we're primarily putting our capital. So I want to talk about the balance sheet. So I had already mentioned at our Q3 call that we will end at $6.5 billion to $7 billion in cash, which is a net cash investment this year of $2.5 billion to $3 billion. So we started the year at $9.5 billion in cash. We are going to have revenue of $1.6 billion to $4 billion or $2 billion. And then we will have cash cost of $4.6 billion, which is a $2.5 billion to $3 billion net cash investment. As I walk that forward, we believe we'll have a net cash investment of roughly $2 billion in 2026. So that is 10% growth on that revenue line, plus an almost $4 billion cash cost, I said $4.2 billion. So that should, in the ballpark, basically deploy $2 billion of capital, in which case, we would have $4.5 billion to $5 billion by the end of 2026. Then as we turn to 2027, we're targeting $1 billion. So we've had a mantra of burning -- investing $3 billion in cash in 2025, $2 billion in cash in 2026, and $1 billion in cash in 2027 before we break even. That is still our mantra. That is still what we're targeting. We've provided here for $1 billion to $1.5 billion because we'll have to see where the revenue line grows. There are all the opportunities that Stephen has already laid out for you, and we'll see what happens. Maybe it's $1 billion, but if it's not exactly at that revenue line yet, maybe it's $1.5 billion of net cash investment income in 2027. So therefore, we'll end at $3 billion to $4 billion in cash, which is a very strong balance sheet before the year that we break even. So we're excited by that. We'll see revenue growth as well as cost reduction that will basically leave us with a strong ending cash balance and a strong balance sheet of $3 billion to $4 billion. So then you might ask why did we then go out and announce an exciting loan this morning? And really, that is a strategic move to maximize flexibility here. So obviously, we're excited about that. We want to make sure that we are in control. And I think if you look at the terms of this loan, we are quite excited about it. Number one, it's non-dilutive financing from an equity perspective. Number two, it's relatively low cost, and let me explain that for a second. So you can see SOFR plus 550. So right now, that's about a 10% interest cost. But we will just take the cash proceeds that we get, which the first draw is $600 million and put it in the bank. So the ultimate interest cost is the 550 million spread. So that's what we'll be sitting on for $600 million of this. And then the delayed draw component is just pure flexibility over the next 2 to 3 years, and that is at a very nominal cost. The average cost of that for 3 years is 1%. So that gives us significant flexibility, which we're excited about. We're still able to use it for any general purposes. So we can use it for business development. We can use it for share repurchases. And this is a 5-year loan. So a lot can happen over 5 years. And that's what gives us extreme flexibility and a lot of comfortability over the next few years. So then just to update our 2025 ending cash balance. That means we will now end the year with $7.1 billion to $7.6 billion in cash because we'll draw $600 million. And our liquidity will be over $8 billion at the end of this year, which we believe on a year that we are investing $2.5 billion to $3 billion, and that investment -- net investment will reduce over the coming years. That gives us a lot of flexibility and a lot of comfort as we look at our financial future. And if you fast forward to 2027, that means our liquidity will be roughly $5 billion. So going into a year that we break even, we believe we'll break even in 2028 on a cash basis, we will have $5 billion in flexibility in liquidity, $3 billion to $4 billion from a cash perspective and then this extra $900 million from a delayed draw loan perspective, which we're really confident in. So overall, we really believe in the base plan, but this provides a lot more flexibility to us moving forward as well. So I want to summarize what I think are the key financial takeaways. I really believe this is a turning point in our financial story. And that turning point starts with growth. So we believe we will grow over the coming years. We've tried to lay that out for you. The first year will be up to 10%, and we believe we will continue to grow thereafter. And I think that's a big change versus the last few years. The second thing is, Jehr laid out a terrific story for gross margin expansion. I would say of this 10%, half of that is volume and the other half is all the productivity drivers that we are driving. So we feel very confident that with revenue growth and with the outstanding efforts of the entire team and what they're driving from a manufacturing efficiencies perspective that we will grow 10-plus percent on the gross margin line. We will evolve our R&D investment from infectious disease, and we will take that down to a life cycle management level by 2028. And then we will significantly invest into oncology and hopefully rare and autoimmune behind that as well. But for the most part, it will be an oncology story over the coming years. And then we're reducing our cash cost and investments. So by 2027, we will be $3.5 billion to $3.9 billion, and that really gives us great comfort in seeing cash breakeven by 2028, great line of sight to that. So we believe in this strong financial framework. We believe that we will have $3 billion to $4 billion in cash absent the loan, but this enhanced liquidity gives us a lot of flexibility for both uncertainties or opportunities that present in our future. So with that, those are the financial takeaways that I'd like to wrap up with, and I will turn it over to Jackie.

Great. Well, good morning, good afternoon and good evening. It's a pleasure really always to be with you again. And this year, I again have the happy task of taking you through our scientific data with the team. And I want to say, unlike Jamey, we're going to talk about many things, which actually is a super pleasure. So you're going to see a wide variety of the team presenting data with me. And I think that also is a representation of our success. We've really built out a large team of investments who are representing, as Jehr said, an amazing team behind all of these data. So with that, why don't we get started? So we're first going to talk about our seasonal vaccine pipeline and some of the new data there. This has obviously been a big year for us in seasonal vaccines. So in the U.S. alone, we've had three FDA approvals, and that's starting with mNEXSPIKE, which, as Stephen explained to you, has been incredibly important to our U.S. business this year. That team has also pulled forward two additional supplemental BLAs, first for RSV. So keeping up with our competition in having data in those high-risk individuals recommended for vaccination 18 to 59. And then in COVID, importantly, the only licensed product currently in the U.S. for pediatric patients. And so you'll see representatives from the entire team who will talk to some of the data that we've generated this year. I'm going to finish the presentation on Norovirus, which those of you who have followed this story, you'll say, wow, Norovirus changed teams. What happened there? We're now thinking as Norovirus gets closer to being a commercial product, Norovirus more and more seems like it's going to share some characteristics with our seasonal vaccine portfolio. So like other seasonal vaccines, it has a period of intense transmission, and that's in the winter season. Like COVID, it also has a small bump in the summer, but we expect that most vaccinations are going to occur in advance of that winter season. And also like COVID and like influenza, it really can vary in terms of epidemiology over time, and we're envisioning a world where maybe not every year, but some years, we may want to update that vaccine. And so we've gained some expertise, obviously, in COVID, learning to do that more with influenza this year and ultimately, we'll apply that to Norovirus. So with that, we're going to start our seasonal journey, and I'm going to hand the podium over to Darin Edwards, who is our program leader on that program.

Thanks, Jackie. So I'm really excited to be here to talk through our data on mRNA-1283, which is mNEXSPIKE as it is branded in the U.S., and we've done -- not only gained the approval, but also done the annual strain update and launched the product in the U.S. So this is data that comes from our pivotal Phase III study, and it's data that we were excited to receive, excited to report, and I'm excited to talk about it today. So a little bit of a description of the trial before I get into the results. So this was a Phase III trial designed to assess the immunogenicity, the safety and the -- relative vaccine efficacy of mRNA-1283 versus our commercial product, mRNA-1273. Approximately 11,500 participants enrolled in this study above the age of 12 years and older. A single dose of either mRNA-1283.222 or 1273.222 was administered, and that is the bivalent vaccine composition encoding the ancestral and the BA.4/5 strain. And this was used because that was the licensed product composition at the time the study was enrolled. So first, a little bit about the participants in this trial. So characterizing the demographics and the baseline characteristics of these groups, they were well-balanced between the two arms. That's both in terms of age with a median age of approximately 56 years for participants in the trial, breakdown between the different age cohorts as well as by race and ethnicity where the demographics were largely representative of the American population. One thing to highlight, though, is approximately 46% of the participants in this trial also had greater than one or greater comorbidities as defined by the CDC case definition. In addition to the demographics, we also controlled for prior SARS-CoV-2 infection with about 75% of participants that actually had evidence of prior SARS-CoV-2 infection, very characteristic of the time during which the trial was enrolled. In addition, most of the participants in this trial had 3 to 4 prior vaccine doses with a median time of about 10 months from the last vaccine dose. So getting into the results from the trial. First, we look at solicited local adverse reactions, and they were relatively well balanced between the two arms with a trend towards lower local reactions for the mRNA-1273 arm. The highest solicited local adverse reaction was for pain. But for all, they typically had either Grade 1 or Grade 2 reactions that resolved after about 1 to 2 days. Now looking at the systemic adverse reactions. We see they were well-balanced between the two arms of the study 1283 and Spikevax. Fatigue, headache, and myalgia were the most commonly reported systemic adverse reactions. Again, Grade 1 or Grade 2 reactions were most common. And again, they resolved in about 1 to 2 days. So this study was designed to assess relative vaccine efficacy. And to do that, we use the CDC COVID-19 definition for COVID-19 symptomatic disease. And that required a virological confirmation of SARS-CoV-2 infection via PCR as well as the presence of one or more symptoms consistent with COVID-19 within 14 days of the positive PCR. Active surveillance was conducted in the study in order to capture cases, and that included biweekly symptom surveillance conducted using our e-diary as well as assessment of participants with symptoms for clinical evaluation and for collection of samples that we then used for PCR. So starting out with the top line data from this study. So the prespecified success criteria was met for relative vaccine efficacy for 1283 versus Spikevax. And this is indicated by the data that you see on the screen, where a 9.3% positive point estimate was seen for 1283 in relation to Spikevax in terms of prevention of symptomatic COVID disease. The success criteria required at the lower bound of the confidence interval to remain above minus 10%, and that's what we see from these results. But as we know, the people at greatest risk for COVID-19 disease are those older adults. Here, we see a breakdown across three different age ranges. And it was very nice to see that the highest point estimate that we saw from the study when we did an age breakdown was for those at highest risk for symptomatic and severe COVID-19 disease, that being the above 65-year-olds. There, we saw a 13.5% positive point estimate in favor of mRNA-1283. As we look at the younger cohort that we see relative consistency between the two arms, but the confidence intervals are very wide based on the limited case number and the limited number of participants in that age range. So I think we all know that it's not just age, but it's also the presence of risk factors or comorbidities that put people at higher risk for severe COVID-19. So we wanted to do a post-hoc analysis of the results of the data from this study in order to give us a view on how 1283 is actually performing in prevention of COVID-19 disease in individuals that had these comorbidities. So the top line in this figure demonstrates or it captures all people in this trial, 12 and older that had one or more comorbidities. And here, we see a 17.5% point estimate for prevention of COVID-19 disease. For those that were above 50 years old, again, increasing the level of risk, increasing -- looking at populations that had increasing risk. So above 50 years old and with one or more comorbidities, we see an even higher point estimate that being 23%. And above the age of 65 with one or more comorbidities, individuals that had the highest risk for severe outcomes, we see a point estimate of 28.6%. We're very happy to see these results. So this study was not powered to assess prevention of hospitalizations, but we were in a post-hoc analysis able to use the FDA-defined guidance, FDA severe disease criteria in order to do a post-hoc analysis to give us some view on how 1283 is actually performing in relation to spikes in prevention of severe disease. Now this is all built on the knowledge that Spikevax has been demonstrated consistently to be very effective in preventing severe COVID-19, not only in our pivotal efficacy trial, but also in real-world effectiveness studies that we conduct year-over-year for each vaccine composition that we gain approval for and launch. We were able to identify 55 cases in this study that met the FDA criteria for severe COVID-19. And in this assessment, we see a 38.1% positive point estimate in favor of mRNA-1283 versus Spikevax. So now pivoting to the immunogenicity results from this trial. At the top, you can see the view of the neutralizing antibody responses against the two variants that were encoded in this vaccine composition, the original SARS-CoV-2 and BA.4/5. For both of those, the non-inferiority criteria was met, that being a lower confidence interval above 0.667. But importantly, you see that the lower bound of the confidence interval is actually above 1. So that indicates that mRNA-1283 is driving a higher immune response. Same thing looking at seroresponse rate, the non-inferiority criteria was met, that being a lower bound of the confidence interval above minus 10%. But again, the lower bound is above 0, again, indicating that 1283 is eliciting a stronger immune reaction than the very strong immune response that we get from -- have measured consistently from Spikevax. Again, we wanted to look at those individuals that were at higher risk. So we did an age breakdown. And in this assessment, you see that the highest GMR that the ratio between the titers elicited by 1283 versus 1273 was highest in the above 65-year-olds, correlates very nicely to the relative vaccine efficacy results that we also measured between -- in those age ranges. We also wanted to take a look at durability of this product in relation to the immune -- the durability profile that we have consistently measured for Spikevax. So here, we're looking at 1 month, 3 months and 8 months -- or sorry, 6 months and looking at not only the GMR, but the overall titers that we actually measure for Spikevax and also for mRNA-1283. And we see the GMRs are very consistent at these three different time points. And also, I think it's actually very important to highlight that the overall neutralizing antibody response that we measure for 1283 at 6 months over 1,000 is actually higher than the neutralizing antibody response that we measure for Spikevax at 3 months. So it's been my pleasure to be up here talking through the pivotal Phase III results that supported our licensure in the U.S. and are also supporting the reviews of our 1283 applications globally. Just a brief recap, mRNA-1283 was generally well-tolerated, had an acceptable safety profile. We met the prespecified relative vaccine efficacy non-inferiority endpoints with a point estimate of 9.3% for all participants included in the trial. We saw a trend for higher RVE point estimates with advancing age and with comorbidities. For all participants above the age of 65, we saw a 13.5% point estimate. And for above the age of 65 with at least one comorbidity, we saw a point estimate of 28.6%. We met all prespecified non-inferiority objectives for immunogenicity. That includes indication that mRNA-1283 elicits a higher immune response than Spikevax. And that higher immune response was most evident in individuals that were older. And we plan -- we have plans and those are actively underway to augment this data with data from clinical studies of our currently approved LP.8.1 vaccine composition, and that data will be published relatively soon. And we're also conducting real-world evidence studies, again, and that data will become available as it accumulates. So we're approved in the U.S., Stephen shared that information, and that is for the current LP.8.1 composition. We're approved in Canada, and we're targeting strain update for next year. And we filed for and are targeting 2026 approvals and strain updates in other markets like Australia, Europe, Japan, and Taiwan. So thank you very much for your time. And I think next, I'll pass it to Raffael to talk about flu.

Good morning. I'm really excited to give an update on our influenza program today. A quick reminder, as Stéphane said, influenza is a major source of morbidity and mortality worldwide. Up to 1 billion cases occur every year and in the U.S. alone, up to 130,000 deaths every single year. What's also important to highlight is that age and chronic conditions are also playing a role in the increased risk for influenza and influenza can actually cause downstream events like cardiac as well as pulmonary events like heart attacks, stroke, or COPD. As you all know, influenza vaccines are licensed. There's also an enhanced vaccine market with vaccines that have shown superior efficacy to the standard dose vaccines. A quick reminder to mRNA-1010. It encodes for the surface glycoproteins, the major immune response for the influenza vaccines, the hemagglutinin. And there are some inherent advantages of our mRNA platform when it comes to influenza vaccines. It encodes the exact protein that is being recommended by WHO and other recommending bodies. It has no requirement to propagate the virus in cell lines or eggs where you can sometimes get mutations that can actually change the antigenicity. And what's also important is downstream, we have reduced production times, which gives the potential of having latest strain recommendations as well, which could improve the matching of the vaccine strains to what is circulating in nature. What I'm not going to show today, but we've previously shown that we actually get superior immune responses of our mRNA-1010 vaccine to both standard as well as enhanced vaccines. But today, I'm actually going to talk about the efficacy that we've seen in our P304 study that we conducted last year in the Northern Hemisphere. It was a large trial of 40,703 participants. We randomized them to either receive mRNA-1010 or a licensed comparator vaccine. Last year, as some of you might recall, was when first the change of removing B/Yamagata happened from the vaccines. Not all regions made the change at the same time. So depending on the region, we actually had a trivalent or quadrivalent comparator. MRNA-1010 was trivalent throughout globally. In terms of the study objectives, the primary objective was to show non-inferiority and superiority of mRNA-1010 versus the standard dose comparator in terms of efficacy against all influenza strains. We also looked at the safety and reactogenicity of mRNA-1010. And then in secondary objectives, we also looked at specific match to strains that were circulating and also the immunogenicity of a subset of participants to confirm the prior findings that we had in our previous studies. In terms of exploratory objectives, we also wanted to look at the impact of mRNA-1010 against medically attended ILIs, which are more rare events. But they are, as I mentioned, really important as well in terms of the types of disease that you actually want to prevent. If we look at the demographics, you see again that it's a well-balanced group between mRNA-1010 and the licensed vaccine comparator. What I do want to highlight is the last line, which is the baseline high-risk conditions. More than half of the participants had some type of baseline high-risk condition, which included diabetes, asthma, obesity as well as COPD and atrial fibrillation. And when we go to first the reactogenicity, it's very much in line with what we previously had presented about mRNA-1010. We do see higher local reactogenicity with the majority being injection site pain, Grade 1 and Grade 2 in nature and transient. Similarly, for the systemic reactogenicity, we do see higher reactogenicity for mRNA-1010 compared to a licensed standard dose comparator. However, a low frequency of Grade 3 events and most reactions were Grade 1 or Grade 2 in nature. The most reported systemic reactions were fatigue, headaches in both of the groups. This is, to me, the most exciting slide of all of it. This is -- Darin has shown a very similar slide for COVID, but just to orient you because we have a couple of more slides. We have the geometric mean is the blue diamond on top, which is the 26.6%, but then we have the confidence intervals. And then you have those different lines that actually show the non-inferiority bar. They show the superiority bar and then the highest superiority bar. And what you can really see is that the lower bound exceeds all of those bars, which really gives us a lot of confidence with those results standing for mRNA-1010 being, in fact, superior to the standard dose vaccine. And this also holds true when we then break it down by strains. The confidence intervals become a bit wider because, of course, you don't have as many cases for every single one of those strains. But you actually see that we get that efficacy for both the influenza A and influenza B strains across, which is really reassuring to see for our vaccine. This I really like a lot because it really drives home that the waning that has been known for seasonal influenza vaccines for years is not a bigger concern for mRNA vaccines. In fact, actually, we see over the season that the separation of the curves becomes bigger with mRNA-1010 actually showing constant protection throughout the season, which is really reassuring to see where you still see an increase of cases later in the season with the licensed standard dose vaccine comparator, you see the curve flattening out for mRNA-1010 and then staying separated throughout the season. So that's really exciting for us to see. A couple more breakdowns, which I think the drive message is that we see consistency across all those different cross-sections that we can make. When we look at the different age groups in this study from 50 to 75 years and older, you see that we consistently got the relative vaccine efficacy relative to the standard dose comparator. If we go to the high-risk conditions, we again see the same picture. We see that frailty status. We see from the fit to the most frail, the consistent higher relative vaccine efficacy of mRNA-1010. We see with the higher BMI that higher relative vaccine efficacy. So that was really nice and reassuring to see. And then as I mentioned, we had an exploratory outcome to look at the medically attended ILIs. And if you go from the top down, you effectively see different levels of severity. It goes from urgent care visit all the way to hospitalization. For the hospitalization ER visits, we didn't have as many cases. They are pretty rare events. But you see that for urgent care visits, outpatient visits and the overall counters, we consistently see that superior relative vaccine efficacy. In terms of the summary, the reactogenicity, as I mentioned, was higher for mRNA-1010. However, most of the solicited reactions were Grade 1 and Grade 2 in nature and transient. We saw an overall acceptable safety profile for mRNA-1010. We saw higher efficacy across all age groups, influenza strains, including participants at high risk of severe influenza compared to a standard dose vaccine comparator. The efficacy was maintained throughout the entire season. And what I didn't show here, but I mentioned that subset of immunogenicity, we yet again saw the same superior immunogenicity as well. And mRNA-1010 prevented more severe and medically attended influenza. And as Stephen alluded to earlier today, we are in the process of submitting filings, and we are intending to submit to the U.S., EU, Canada, and Australia by January 2026. And with that, I'm handing over to Christy to talk about the combination vaccines.

Hello, everyone. So Christine Shaw, I'm the Portfolio Head for Infectious Disease in Rare. And today, I'm going to first share an update on our combination vaccine against flu and COVID. And this combination vaccine takes into the two components that you just heard Darin and Raffael talk about, our 1010 vaccine for flu and our 1283 vaccine for COVID-19. So despite available vaccines, commercial vaccines for both COVID and flu, there remains a high unmet need and burden of hospitalizations for these diseases. And this visual on the left shows in the U.S. in adults above 75 years, the amount of hospitalizations over the last two seasons for COVID in red and flu in blue. And one of the reasons for the still significant burden is the low vaccine coverage rate, particularly for COVID-19. And you can see on the right that, that coverage rate in the 75-plus population is around 40% in the last 2 years. As we know, the flu vaccine coverage rate is a bit higher at 75%. And as Stephen shared earlier this morning, the individuals getting a flu vaccine, about 30% to 50% depending on the year in this age group do get a COVID vaccine at the same time. So we think that a combination vaccine against flu and COVID can do two things. One is it can help to increase the COVID vaccine coverage because individuals coming in to get their flu shot can now get a combination shot in one arm and one injection that covers both diseases, both pathogens. And it will help for the convenience of those already getting both vaccines, they can get it in one shot instead of two. So overall, this should reduce the burden of disease against these two different pathogens. So our 1083 vaccine, as I said, is a combination of the 1010 flu and the 1283 COVID vaccine. And we've previously shared that the Phase III pivotal study for this vaccine was successful and met the primary endpoints. So I'm going to review that study briefly and also share some of the durability antibody responses in the study. So it's a randomized observer-blind active control study in which we assessed safety, reacto and immunogenicity. The study is split into two age cohorts. The Cohort A is above 65 years. Cohort B is 50 to 64 years. And the reason we did this two cohort is that we are able to compare the 1083 vaccine to co-administered licensed comparator flu and COVID vaccines. And in the 65 plus, we can do that compared to Fluzone HD, which is an enhanced flu product. And in the younger age cohort, the flu comparator is Fluarix, which is a standard dose vaccine. This is the standard of care for flu vaccines in the United States. In both cohorts, the COVID vaccine comparator was Spikevax. And there's about 2,000 participants in each of these cohorts. So 1083 showed an acceptable reactogenicity profile in both of the cohorts. We have a majority of the reactions were Grade 2 -- Grade 1 and 2 in severity. There was a somewhat higher reactogenicity profile for 1083 relative to the co-administered licensed comparator vaccines. And the reactogenicity was a bit lower in the 65 and older cohort compared to the younger cohort. So here are the immunogenicity data. And again, we assess this independently in the two age cohorts. In each cohort, what we did is determine the ratio of the antibody response to 1083 compared to the antibody response to the licensed comparator vaccines. So for flu, each of the four strains in the vaccine, we did this by hemagglutinin inhibition assay or HAI assay. And those data are shown in blue. And for SARS-CoV-2, we measured neutralizing antibody and those data are shown in red. And you can see for all of the strains in the vaccine, the antibody response met the predefined success non-inferiority criteria and that the lower bound of a confidence interval was above 0.667. But we actually saw a higher response to the vaccine for the flu antigens, H1 -- flu strains H1, H3 and B Victoria. And these are the clinically relevant strains, as Raffael shared, B/Yamagata is no longer circulating and therefore, not recommended to be part of seasonal vaccines going forward. So in these cases, the response even above Fluzone HD was higher and that the lower bound was above 1. We saw similar results and just as importantly, against the COVID -- antibodies against COVID-19 or against SARS-CoV-2, sorry, in which the responses to 1083 are higher than the responses to Spikevax. And this was true in both age cohorts. So the data -- new data sharing today is looking at the durability of that antibody response. So we took samples 6 months after vaccination and measured the antibody response in the same assays. And you can see in the top panel are the older adults and in the bottom panel are those below 65. And the 1083 groups are in red. And you can see that throughout the time course in the study, the response to 1083 is higher or equal to that of the comparator vaccine out to the 6-month time point. So to conclude this part of the 1083 showed an acceptable safety and reactogenicity profile compared to the comparator vaccines. It did meet all the co-primary immunogenicity endpoints and elicited a higher immune response against SARS-CoV-2 and the flu strains -- the relevant flu strains in both age groups in the study. And because antibodies are an established surrogate of protection against both flu and COVID and because both components of the vaccine have demonstrated efficacy in stand-alone pivotal efficacy studies, we think that the combination vaccine 1083 should also protect individuals from both COVID and flu. So as noted earlier in the presentations today, the filing for this 1083 vaccine is under licensure review by EMA. We also recently submitted application to Canada, and we are waiting further guidance from FDA before refiling there. Overall, we're very optimistic that this vaccine can help reduce the burden of disease against both flu and COVID in the coming years. So switching gears to RSV, mRNA-1345. We are going to share a bit of an update on that vaccine as well. All right. So RSV still causes significant burden of disease in the U.S. despite there being vaccines. And in this past year, you can see here that there are millions of inpatient visits, outpatient visits due to RSV. There's hundreds of thousands of hospitalizations and tens of thousands of RSV-related deaths. And we know that vaccination is an effective strategy to reduce this disease burden. Our mRESVIA-1345 vaccine was licensed initially in adults 60 years and above. And at this point, we've achieved licensure in 40 countries across the world, as shown on this slide here. Recently, we have achieved expanded indication licensure in another age group in U.S. and in Europe for individuals 18 to 59 years of age with underlying comorbidities. Today, what I'm going to share on this program are the revaccination results, both safety and immunogenicity from two studies, one in which we gave a 1-year revaccination and the other, which we gave a 2-year revaccination. And so first, the study with the 1-year revax. So this was a Phase III study of P302. And in this study, individuals who had received mRESVIA vaccine at day 1 received a revaccination about 1 year later with the same commercial dose level. We measured safety, reactogenicity, and non-inferiority of the immune response after the revaccination compared to the antibody levels after the first day 1 dose. So here's a visual of the participants in the study. This was a study in 50 years and above and about 543 participants were enrolled in the study with a -- having about 60% of them above 60 years of age and a bit more than half of the individuals female. So this revaccination was well tolerated, as you can see in the visual here, with local and systemic reactions primarily Grade 1 or 2, a median onset of 1 to 2 days and a median duration of 2 days. And now looking at the antibody results from this study. At the 12-month post initial dose time point, you can see there still are measurable antibodies that are above baseline, about twofold above the baseline day 1. And when these individuals got a revaccination at that 12-month time point, the antibodies were restored up to the level seen after the primary injection, where we had demonstrated efficacy previously. So this study also met the predefined non-inferiority success criteria, both against RSV-A, which is shown here as well as RSV-B not shown on this slide. And then moving to the second study in which we now assessed revaccination after a 2-year period. And this study was actually part of our pivotal Phase III efficacy study, and we took a subset of the individuals and gave them a second injection, 2 years after their first injection. They were randomized to receive either the same commercial dose of mRESVIA or a placebo injection. And again, it was a safety immunogenicity study and the endpoints were similar to that I just described for the 12-month booster study. This study was in 60 years and above and about 1,000 of them received the mRESVIA vaccine about 2 years after their first dose, as I mentioned. The median age here is 68 years and about 50% female. And here, were about 1/3 of the participants had an underlying comorbidity and 1/3 of the participants were defined as obese. So I'm sharing the reactogenicity data split across two slides with this first slide being the local reactogenicity. And most of the events were Grade 1 with median onset of 2 days and duration of 2 and overall showing that this revaccination at 24 months was well tolerated. And now the systemic adverse reactions showing a similar picture where the reactions were mostly Grade 1 and 2 with an onset of day 2 and a short duration, again, showing well-tolerated revaccination. So the immunogenicity after this 2-year booster is on this visual here, showing the RSV-A neutralizing antibody responses. And similar to what I showed you with the 12-month booster here, once you go out to 24 months, you again, still see antibodies restored above the -- or maintained above the baseline after the first injection. So you can see it's about 4,000 versus 2,000, so about a twofold elevation still, which is actually quite similar to the level we saw in the 12-month post vaccination in the previous study. And again, after the revaccination at this 24-month time point, the antibody titers were restored and reached a level similar to that after the primary injection. And in this study, we also achieved the predefined non-inferiority success criteria, both for RSV-A, the visual here and for RSV-B, which is not shown. So summarizing the RSV story shared today, revaccination either at 12 or 24 months is well tolerated. And we are able to show that durability of the immune response does last out to at least 2 years. And if you give a revaccination at 12 or 24 months, it does restore the immune response and it does meet non-inferiority success criteria. So we think because of these results that revaccination at either of these time points would be expected to provide comparable vaccine efficacy to that after a primary dose. And therefore, revaccination has the potential to provide sustained protection against RSV, particularly in individuals with underlying comorbidities in which revaccination might be particularly beneficial. So we continue to monitor guidance from recommending bodies on RSV revaccination approach and timing. Turning back to Jackie for Norovirus.

All right. Well, thank you, Christy and team. And I think you can see what a pleasure it is to work with this group of people every day. So now I'm going to talk about our newest seasonal vaccine, Norovirus and our progress there. And just as a reminder about why we think Norovirus can be such a growth driver for us. So among enteric viruses, this is really the leading cause of diarrheal disease globally, and it results in a substantial healthcare burden. And I'll just say now that rotavirus is a vaccine preventable disease in children. This is a virus that actually can impact both older adults and children. Unlike rotovirus, Norovirus actually has its most severe disease in older adults and immunocompromised patients. So while highest incidence is in kids, greatest impact in terms of severity of disease is in older adults and immunocompromised patients. And that really defines why we've designed our clinical development program the way we have, which I'll talk about in a couple of slides. The burden amongst older adults is also expected to rise along with societal aging and the increased need for institutionalized care. So this is one of those viruses just like respiratory viruses. And if you have a contained group of individuals, it's a fecal oral virus, meaning it's passed through inappropriate hand hygiene. You can imagine that where care is being given, it can just rip through one of those institutions. So we think it's going to be important in those settings. And what you see in the United States, about 20 million infections leading to about 900 deaths, but 100,000 hospitalizations where people will receive supportive care, so rehydration primarily and then also support for maybe organs that have been impacted by reduced perfusion. That's $2 billion of healthcare costs annually and lost productivity. Same numbers for the global environment, about 685 million infections each year with 200,000 deaths and about $60 billion lost in terms of healthcare costs and lost productivity. So I mentioned that Norovirus shares some attributes with other seasonal viruses. And really, it's the variability that's one of the key pieces. And I alluded to this a bit earlier. But just to say the reason why you will see us pursuing a multivalent vaccine is because there's actually quite a lot of distribution of genome groups. We've targeted the genotypes that are most prevalent year-on-year. So we've really looked across the last couple of decades to design the first version of the vaccine. They're actually both trivalent and pentavalent versions. We're going forward with the trivalent first. If that's successful, the pentavalent in pediatrics, where we actually see the greatest distribution of genome groups will be how we move forward next. And an important piece about the mRNA technology and why we think this is an ideal technology for norovirus. As I mentioned, with the genetic diversity of these viruses, the ability to have a second option in terms of strategies for greater vaccine coverage, meaning rather than continually adding different strains, the way we've done with pneumococcal vaccines, which leads to increased immune interference also can lead to challenges in terms of how many can you actually put in that syringe. Every iteration of pneumococcal vaccine gets further away in development because it's harder to both develop and manufacture. With the mRNA technology, as we do surveillance, you can actually swap in and out strains as they become relevant or recede in relevance. And this is actually a discussion we've already started with the FDA, thinking about a diarrheal vaccine maybe differently than people have thought about it before. So I'm going to share with you our Phase I clinical trial. We actually investigated both formulations, the trivalent, which is 1403 and pentavalent 1405 for obvious reasons. But we observed them in the same clinical trial so we could pick that formulation we were going to take forward into Phase III in adults. It was a randomized, observer-blind, placebo-controlled trial. And it actually had 664 healthy volunteers, pretty big for a Phase I study, but that's because of all of the treatment groups, so multiple dose levels in two formulations. We looked at both one and two doses. Why? In children who are primarily naive to the infection, we think we may need more than one dose to get to protection. But in adults, because all of us have experienced Norovirus at some point, one dose is really going to be sufficient. And that's how we've gone forward into Phase III, and those are the data I'll really be sharing. We've been following participants for 12 months after their last study injection, and that's really to look at the persistence of those antibodies to think about how long could this last? And I'll be sharing some of that with you today. And this was primarily only a U.S. study. So before we get into the immunogenicity results, I think sometimes it's important to really understand what we're measuring when we do these immunological assays and maybe to also help you understand why we think we have an understanding of how this vaccine might work and what makes us comfortable to make that next stage of investment. So the assay that we've been primarily looking at is a functional assay, meaning we're not only measuring how much antibody we're making, we're also observing how that antibody interacts with the pathogen and -- or with the human body to understand maybe how it might work. In this case, the histo-blood group antigen blocking assay and from now on, I'm going to say HBGA because it's a lot easier. HBGA is an antigen that's really important for the virus entry into those epithelial cells in the gut. And what we're looking to do is blockade that virus entry. So this is really a binding antibody. And I -- for the sake of time, I'm not going to go through the details of how the assay works. But just to emphasize, we're looking both at how much antibody we induce and also the quality of that antibody. So we looked in two cohorts, both older adults and younger adults. So on this slide, you see the older adults. And as I mentioned, I think the target for this population is obvious. It's really because they experience the most severe disease. As we age, we're more susceptible to this kind of disease. And as people age, they're more likely to be in settings that really favor outbreaks of this disease. So what you see here are the three genotypes that are included in our vaccine. And you see the results from day 1 and day 29. Remember, I told you I'm going to share with you the single-dose data because that's really how we envision this going forward, just like getting your single flu vaccine every year or in the case of Norovirus, perhaps every other year, every third year remains to be seen like with RSV. What we see is some seropositivity, and it actually really varies across the different genotypes as you might see at the day 1 level. And why? Interestingly, Norovirus, some of these genotypes actually tend to evolve much more quickly than other genotypes where they tend to be more stable. And so where you see lower pre-vaccination titers -- that's in the geno group GII.4, where we see the most variability. So understanding even whether we ultimately may be able to swap out GII.4 strains might be helpful in the future. But what we see with all three of our genotypes regardless of the pre-vaccination titers is a really robust increase in antibodies and really going from 10 to the 2 to really 10 to the 3, close to 10 to the 4. So we're really encouraged by these results, again, in this functional assay, looking at how the antibodies are able to blockade that key cell surface antigen for viral entry. And now here, you see the data in younger adults, and it's really a similar story. Why younger adults? Well, younger adults, first of all, may be in occupations where they're really at high risk. And so just like this virus can rip through assisted living facilities, it definitely rips through pediatric and adult hospitals every year. And if you are a healthcare provider who's being exposed to that, there's pretty high force of infection. There are other populations that may choose to be vaccinated as well, like, for example, parents of young children, if you've ever taken care of your child and then experience the infection after your child, I used to be a mom whose kid like to climb in my lap and throw up, never in the toilet. This is a virus that I would sign up to be vaccinated against. I'll put it that way. So safety data, again, as we show in all of our slides, dark blue are Grade 1, lighter blue are Grade 2 and the orange are Grade 3. We see our typical pattern in terms of the local reactions. And in the different dose levels, we see in younger adults increasing rates with increasing doses. What you'll see is, unlike some of our other vaccines, that band of orange is actually really small. So we're very encouraged actually by the reactogenicity profile. And this is something we see. There's generally an increase with increasing doses regardless of the vaccine antigen that's part of the platform, but the different antigens cause different levels of reactogenicity. Older adults, less reactogenic than younger adults, and that's something we've also typically seen. So as we talk about now, we've moved to Phase III, and I mentioned to you the reason why we're really going forward with trivalent in adults, but with a longer view towards if this works, moving to pentavalent and starting in the pediatric population. So this is one of our large-scale trials, again, placebo-controlled and enrolled about 17,500 per arm. As Jamey mentioned earlier, we did do an initial Northern and Southern Hemisphere season. We're accruing some cases. The epidemiology was not in our favor, unfortunately, this year. So we're going to need another season to accrue additional vaccine-matched cases leading to an interim analysis that we're anticipating later in 2026. So in summary, our single dose was really well-tolerated and showed an acceptable safety profile. The functional antibodies really showed robust titer increases against the vaccine matched strains and the single -- similar mRNA-induced titers were observed in both younger and the older adult populations, which is why our Phase III is actually evaluating the vaccine in those 18 and over. We're advancing now into the third cohort and in fact, have enrolled our first subjects in the U.S. And the Phase III readout is really going to be subject to those case accruals. But as I mentioned, we are anticipating later in 2026, at least an interim analysis. Okay. And with that, I'm actually going to announce a pivot. We pivot a lot at Moderna. We're pivoting to have our break a bit early. So unfortunately, for you, you're going to hear about latent vaccines after. And then I will introduce my colleagues who will lead into the oncology portfolio. So actually, I should say, unfortunately for me because I have to talk to everyone after a break and the insulin effect sets in. But we'll take how long, Lavina? 10-minute break, and I'll see you back then on stage. Thank you so much for your attention. [Break]

Hi, everyone. So thrilled that there are so many great conversations going on, which hopefully we can continue into the lunch and beyond. But if I can ask, I know there are actually a few people still outside, maybe in the next 60 seconds or so, we'll get restarted just to get us back on track. Right. So I hope you enjoyed your break. I'm going to now take you through a tour of our vaccines that are earlier in development before handing it over to the oncology team. So what you see here are the latent and bacterial vaccines. We're going to talk for a minute about CMV vaccine, followed by our two EBV candidates, the prophylactic candidate and the therapeutic candidate. And then I'll end with a quick talk on Lyme vaccines, and then I'm going to hand it over then to my colleagues who are really leading the charge in oncology. All right. So CMV in transplant was obviously a huge disappointment to us that we were unable to demonstrate prevention of infection in seronegative subjects. This was obviously our first step to really getting to the indication of prevention and congenital infection. But I will say it was always a very high-risk program in the sense that we were studying seronegative patients and hoping to prevent evidence of any infection. And as you may remember from our COVID in our RSV Phase III trials, we were able to show some prevention of infection, but really at much lower rates than symptomatic infection or severe disease. And that's pretty typical, not just in the vaccine world, but in general, it's -- you often see evidence of more severe disease being the first place that you have impact in infectious disease. So why do we believe in CMV transplant? Well, one, it's a very different population. So while there can be CMV negative transplant recipients, the reality is, as you get older, you are more likely to have a disease that leads to a failure of some end organ or cancer. And that means that transplants are actually weighted more at the older end of the age spectrum. So these people tend to be seropositive and the pathophysiology of their disease, either because they're receiving a solid organ that has CMV-infected cells or because they themselves are already seropositive. Once they receive their immunosuppression, that really reduces the immune control of latency of that virus and the virus expands. And it can actually cause an immunocompromised people quite symptomatic infection. So the risks that are associated with CMV disease in solid organ transplants and those with hemopoietic stem cell transplants or HCT, graft rejection. So these CMV can actually cause a degree of inflammation and then immune destruction in organs. And since these organs are so very precious, we don't have enough for all the patients that need transplant. That's a huge issue in transplant medicine. Second is it can cause other kinds of end organ CMV disease. One of the most common in this population is CMV enteritis, so causing a lot of diarrhea, potentially leading to dehydration. But you can also see rare forms like CMV retinitis, which is a threat to someone's vision. So there are currently no approved vaccines. There are antivirals, and they're actually used pretty universally. Different centers use them differently, and we'll talk about that in a minute. But the issue with them is, one, they have high cost and they have their own side effects. They're typically not used forever in people with transplants. And what we often see is once you pull off the antiviral suppression, you do see a rebound even when in CMV viremia, so the measurement of CMV circulating in the blood, particularly when the medication is first withdrawn. And that's not a risk that actually decreases. There's that rebound no matter how long you keep that prophylaxis on board if you then pull it off. There are 47,000 organ transplants in the U.S. each year and 23,000 bone marrow transplants. And so that really leads to about 70,000 transplants in the U.S. each year. So a smaller but still sizable market. So sterilizing immunity, as I mentioned, is really challenging in vaccine development. And that's actually not what we're aiming for in the transplant population. As I mentioned, these are typically seropositive individuals. Once you're seropositive, you remain seropositive for life. What we are trying to do is prevent that virus from reactivating in the immunosuppressed milieu. And the way that actually transplant physicians measure whether or not once we withdraw or even when you're on CMV prophylaxis, whether that virus is reactivating is by looking for CMV DNAemia. We actually looked at CMV DNAemia in the Phase III trial in healthy women. And we actually did see that initially after infection, we're able to reduce that level of DNAemia. So that actually gives us hope that this may be a place where we can see some effect. And I'm going to show you what T-cell responses look like in transplant patients in just a moment. So we think it has the possibility to prevent the viral replication and reactivation. And so our hope is that we can look in a smaller bone marrow transplant population. And then if that looks positive, so limiting the investment upfront, but if that looks positive, talk about how we might expand use. So in summary, the CMV is a disease that's associated with substantial morbidity and mortality. It's one of the key issues that transplant physicians across the spectrum have to manage. Letermovir is actually a suppressant that is approved and it's currently being used as standard CMV prophylaxis. So it's really considered best-in-class. And it's a way that you'll see in the next slide, we're considering that as we think about how this program could fit in with existing treatment. And despite its efficacy, it's also been associated with decreased CMV-specific T-cell reconstitution in bone marrow transplant patients. So as I mentioned, these drugs, like all drugs and vaccines have their own side effects associated to them. And it also can lead to late onset CMV infections. Again, once you decide to withdraw your prophylaxis, we often see a rebound reactivation. So that's why we think the development of a safe and effective vaccine that you might be able to boost year-on-year might be a way to get on top of this remaining unmet medical need. So we're studying CMV-positive adults who are over 18 years of age who received an allogeneic, meaning it's coming from themselves, stem cell transplant. We're looking at a slightly different dose level and slightly different schedule than we were in the primary disease. And so this is an accelerated vaccination schedule where we're giving three doses, and that's happening after the initial reconstitution of the immune system. So people are on their letermovir prophylaxis up to day 100. We then start vaccination as they are reconstituting that immune system. And then 6 months into the trial, we're going to give a booster dose to see what it would look like if you needed to boost this over time. There's an mRNA-1647 group and a placebo group and the randomized 1:1. In terms of the solicited adverse reactions, here, you see Grade 1 in gray, Grade 2 in blue, and Grade 3 in orange. And as we might expect actually in a transplant immunosuppressed population, we see lower severity for most of the events, and that's really because probably their bodies are not allowing them to have the same reaction to the vaccine that we see in healthy individuals. However, I was actually pleased to see that we do see some reactogenicity, meaning we're causing their immune system to do something in this case. And now I want to show you actually what the T-cell responses look like. And why the T-cells? T-cells are incredibly important in suppressing CMV and particularly CD8 T-cells are important in controlling that infection and keeping it in latent mode. So what I'm showing you here are the 3 antigens. GB is the antigen that was on its own, gH and gL and UL, they are part of the combination that makes up pentamer. And we're looking at T-cell responses to each of those individual antigens. We also have the CD4 at the top in the different shades of pink, CD8 in the different shades of blue on the bottom. And also what we see are the placebo versus the mRNA. And this is really looking after multiple doses in the schedule. So as I explained to you, we're studying this in stem cell transplant patients. These are people who need to have their bone marrow and their cancer, importantly, completely ablated before we give them a transplant, and that transplant is actually like a seed that grows into a reconstituted immune system. So we're talking about people who are immunosuppressed not only because they're on immunosuppressing drugs to prevent graft-versus-host disease, but because they're in the process of reconstituting that immune system. In the different shades, what you see are the level of polyfunctionality. Polyfunctionality really refers to the different kinds of cytokines and different kinds of activation and cell activity functions that you can see. So darker means better and more functionality. And so what you see at the top are the CD4 responses. I mean it's absolutely clear that there's many more in the 1647 group than we're seeing in that placebo group. And we're seeing about 1/3, 1/3, 1/3 in terms of polyfunctionality. So even at this early stage, we're seeing really reconstitution of a diverse CD4 response. And CD8, similarly, we're seeing polyfunctionality developing. We always see lower CD8s. The fact that we detect them at such a different level than in the placebo group is really something that we feel strongly positive about, particularly in the gB space and in the gL-UL. And what you're seeing in the placebo group there, as you see, it's kind of even throughout. So people, even as they're reconstituting have some kind of baseline level. And this probably is because we're needing some kind of activation, so they're getting a natural boost because they probably are having to control their CMV infection even while on letermovir. So in summary, the solicited local and systemic adverse reactions were mostly Grade 1 and 2. We didn't see any Grade 4 adverse reactions reported. And as I said, in a transplant population, I was actually encouraged to see that we can get them to stimulate some of those reactions. For 1647, the most common were injection-site pain and headache, fever and fatigue, similar to the rest of the platform. Our interim analysis in this population demonstrated that we can induce antigen-specific polyfunctional CD4 and CD8 T-cell responses. And as I mentioned, we think that's encouraging for controlling that DNAemia. Of note, our robust cell-mediated immune responses were observed as early as 77 days after transplant. So this is really speaking to the ability of this platform to induce the right kind of immune responses at a moment when someone really is at their worst in terms of immunogenicity. So we're hopeful about that. And our next step is really to plan our Phase II data readout. So on the heels of our recent CMV findings, we'll be meeting with the MGH team to really talk about how we build in an interim analysis and look at what's happening in this study in the future. Okay. So now I'm going to move on to Epstein-Barr virus or EBV and our prophylactic vaccine 1189 and our therapeutic treatment, 1195. So these two vaccines are really meant to address very different populations. But all of us, just like with CMV, tend to acquire an EBV infection over time. So as you get older, you're more likely to be exposed to this infection and become seropositive. Unlike CMV, there is a very obvious symptom associated with this -- the severe form of this infection. It's infectious mononucleosis. What you see on the left here are data that we actually generated within our epidemiology team, looking at the increase of seroprevalence with age depending on your location. And so what we know is that in high-income countries, you're a little bit slower to acquire the infection than in medium-income countries. But what we were kind of surprised to find in the middle is that by the time you're getting to middle school, about 9 to 11, 12 to 14 years of age, you're looking now at more than half of individuals are actually seropositive. And so this really indicated to us, we need to think about this as a vaccine that we would include with some of the other middle school vaccines like meningococcal vaccine and pertussis. And then finally, I think really pertinent to the 1195 therapeutic program, EBV accounts for over 90% of cases of infectious mononucleosis, so this is importantly severe infection we can monitor and manage and the vast majority are going to be due to EBV. The annual incidence of infectious mononucleosis in the general population is at least 45 per 100,000. So there's a way that we can measure this in Phase III with the peak incidence of about 15 to 19 years of age. So again, this is a program where we're really looking to establish ourselves and begin middle school, high school vaccination programs. The reason why I say this is important for 1195 comes on the next slide, which is if you look at the conditions that EBV can be associated with long term in terms of sequelae, multiple sclerosis tops the list. And it's really when you get to having infectious mononucleosis, if you had the severe form of the infection, your risk is increased even above getting the infection itself. So this is the interplay between the two programs. And I will say our target indication initially is multiple sclerosis because of what you see in the top bar. But we know that Epstein–-Barr, which its latent infection tends to hide in B-cells, is actually associated with a large number of cancers, particularly B-cell lymphomas. And so being able, just like in the CMV program to prevent that vaccine once infection is established to stay in its latent phase, prevent the lytic phase where it reactivates and begins to spread, we think is going to be incredibly important. Okay. So let's talk for a moment about that etiological link that I just referred to. Nearly 1 million people in the U.S. have multiple sclerosis. And this is a progressive degenerative neurologic disease. So people really are on all ranges of the spectrum. I think an important piece is when you're born, you have all the neurons you're ever going to have. So as you grow, you continue to myelinate your nerves, they continue to spread, but you are born with the cells you are going to have. And then once those cells die, you don't get to replace them. Every neuron is precious. And so really being able to have an impact early in disease is incredibly important, and that will become important as we talk about the Phase II study we're currently enrolling. But there was a landmark study that demonstrated that there was a 32-fold increase of developing MS once your sero convert to EBV. This is one of the strongest links ever established between a virus and the potential sequelae happening years later. And it was also previously established. We've talked about this in the past that infectious mononucleosis is a risk factor. So -- and you can see on the bottom that the seroconverted have that 32% increased risk. And then what you see is that on the right, the risk of EBV plus a history of infectious mononucleosis is really increasing up to 2.3 fold over that history of EBV. So that's really an important population and why we think preventing infectious mononucleosis can be important in the prevention also of sequelae. All right. So EBV, as we mentioned, associated with several serious medical conditions that we think can be addressed through mRNA therapies. We talked about IM and multiple sclerosis. I mentioned to you that we believe B-cell type lymphomas, of which post-transplant lymphoproliferative disease is one of them, is another place. I think it's important to also emphasize that in addition to prophylactic oncology, autoimmune diseases, there's a growing body of evidence, and there was a recent paper in science looking at the link of EBV and SLE or systemic lupus erythematosis. So that's a potential place in the future. And there actually is a chronic infection with EBV that's quite debilitating for patients. So these are all places that if we can see impact with our vaccine and therapy, we can look forward to further data generation. So I mentioned to you, we have these two programs, and I wanted to go through with you a little bit the difference between them. 1189 is primarily looking at lytic antigens. Lytic antigens are the ones that are responsible for rapid proliferation and cell-to-cell transfer. 1195 or the therapeutic version also includes those lytic antigens plus two latent antigens. And that's really important because, as I mentioned to you, our goal in these therapies is to not allow that latent virus to get to the lytic space. The virus actually in its latency expresses very different antigens than it expresses in the lytic phase. So only vaccinating against lytic antigens is unlikely to have the impact of keeping that virus suppressed. So that's really why we think we need to look in two distinct spaces. So I'm going to talk about the prophylactic vaccine first. So this was another larger phase trial design. We started in adults 18 to 30 years of age. We then moved down to adolescents 12 to 17 years of age, and that was really based on the epidemiology work we did where we realized we need a younger age group if we're really going to target the majority of infection. It actually incorporated multiple primary objectives. So looking at safety and reactogenicity, of course, but also looking at binding antibodies, so how much antibody we produce and then B-cell neutralization antibodies. Remember, I told you that B-cells are the bad actors in this infection. And so understanding can we neutralize EBV-infected B-cells is an important endpoint as well. We're also going to be looking at epithelial cell neutralization and the impact on viral shedding as well as seroconversion. And if we have cases of mononucleosis, we're looking towards those as well. There's also a Phase II trial that's ongoing. So in the Phase I trial, there was a request from FDA to pause enrollment at one moment and look at safety data. So we have single dose and 2-dose data in 12- to 17-year-olds, and I think that's going to be helpful as we try to decide how many doses of this vaccine do you need. But again, as data continued to accumulate from that epidemiology study, we actually started again looking at an even younger age group, thinking we want to capture as many infections as possible. At this moment, too, we were able to focus our dose ranging at lower doses, and that's really because we're not seeing so many differences in immunogenicity, but this is a way, as I mentioned earlier, for us to manage reactogenicity. Okay. So I'm not going to go through the data in the interest of time, so we can move to oncology, and you've seen it before, but the vaccine was generally well-tolerated in adults and adolescents. Participants showed an increase in functional binding antibodies as well as they showed baseline EBV seropositive threshold. So meaning we can induce antibody titers greater in those seronegative by vaccination than people experienced through natural infection at baseline. And then we also reduced measurable viral DNA and the frequency of shedding in EBV seropositive subjects, meaning we can suppress reactivation and shedding in that case in the seropositives. So the data from Part A and B, meaning those 12- to 17-year-olds are expected later in 2026, and we will have those Phase II data in 2026 as well, hopefully helping us target how we take this program forward. Moving quickly to 1195. As I mentioned, our primary indication is going to be in multiple sclerosis. That's a disease really characterized by immune dysregulation of EBV-infected B-cells presumably by EBV. And this may be one of the key underlying mechanisms, not only of disease initiation, but also of progression, poor EBV control over time. The vaccine mechanism of action is hypothesized to be restoring some robust immune control. Okay. So talking about the Part 1 trial design, this again, was an EBV seropositive subjects only. And again, we're targeting people that are already infected. So once you're infected, we can't make you uninfected. But hopefully, what we can do is prevent some of the long-term sequelae of being infected. We had two formulations and multiple dose levels. Once again, we're really looking at humoral immunogenicity T-cells also being important here, as I mentioned before, because latency really requires the help of T-cells to keep those B-cells in check. Looking at the reactogenicity, here, again, you see the dark blue, medium blue, and orange schema. We actually saw a relatively comparable reactogenicity across the different dose levels. So really are looking to select the optimization between as minimal a dose as possible while seeing the best possible immune responses. Here, you see the binding antibodies. So we have generated binding antibodies to the glycoproteins that are the lytic antigens. Remember, I told you that 1189 and 1195 overlap in this space. And what we see is that in all of the dose levels, even out to day 317, so they finished their dosing schedule. It's a 3-dose schedule at day 180, so seeing 6-month persistence. They're staying above that dotted line, which is the natural infection level. And now I mentioned to you that neutralization is important here, particularly in the B-cells. And again, staying above the limit of quantitation of the assay that's in the gray line. That's true for the B-cell neutralizing antibodies. It's also true for epithelial cells, which can also harbor a latent EBV infection. And then finally, I wanted to share with you the shedding data. So what you see here at the top, I think, in particular, is the placebo group. So these are all EBV-positive subjects. The placebo group is shedding EBV at a certain level over time. And both 1189 and 1195 are reducing that viral shedding over time. So again, hopeful that we can have an impact in that lytic phase of the virus. And then finally, I mentioned to you that particularly for 1195, we think that these T-cell responses are going to be important. At the top, EBNA3A, LMP2B, these are those latent antigens I was referring to where we're going to need to exercise some immune control over those antigens in the latent phase of the virus if we want to prevent that proliferation. So seeing in the pale colors, the median responses being so much higher than the median responses, which are represented by the dark bar in those that received the placebo group, super encouraging to us. And just to say, we also see CD8 responses to the glycoprotein gH as well. And then these are the CD4 responses to the EBNA3 and the gH antigen. And again, including 1189 actually on this -- no, excuse me, the 4 dose levels on this slide, we see CD4 responses as well, both to the latent antigen, EBNA3A and to the lytic antigen gH. So we are so excited that earlier this year, we started enrolling actually now in multiple sclerosis patients. And again, we're repeating the dose-ranging study for two reasons. One, the safety here is going to be important. So these are subjects whose primary disease is neurologic. And as I mentioned, every neuron is precious. So we have an expert DSMB really looking at the progression of those patients over time clinically. We're complementing that with looking at MRIs over time. So one of the key clinical endpoints in EBV is a loss of myelination in the white matter. This is a way that we can follow very early patients over time. This is known to be a clinical endpoint that actually precedes progression of symptoms. And so we'll see between placebo and those receiving 1 of 3 doses, if we can reduce the number of lesions over time. Okay. I think I'm going to go to the summary and for the sake of time. The interim analysis data demonstrate that 1195 is generally well tolerated. EBV seropositive participants show increase in B-cell neutralizing antibodies. The vaccine is able to boost both CD4 and CD8 cells and the humoral and cell-mediated immunity persisted out to day 317, so about 6 months after that last injection. We reduced measurable viral shedding of saliva. It was actually both 1,195 and 1,189. And then looking this year, we have a Phase I Part B data coming in younger subjects later this year. We also have the Phase II multiple sclerosis study that's coming soon. And with that, we're going to move to Lyme disease. So this is our bacterial vaccine program. You all have heard of Lyme disease, I know, especially up here in the Northeast. Lyme disease is the most common vector-borne disease, meaning an insect is carrying it in the Northern Hemisphere. The Lyme follows a bimodal age distribution. So it typically affects children under 15. They're out in the grass playing in the summer and older adults. And interestingly, this is also a seasonal infection, but the season is reversed from those viruses we were talking about. It really tends to proliferate in the summer months when everybody is outside and enjoying being in the great outdoors. In the major geographies, there are about 475,000 cases in the U.S. each year and about 200,000 in the EU. And the symptomatic infection is actually one of the harder ones I found in pediatric practice to diagnose. It can masquerade as a lot of different things, but subjects typically develop a rash. That rash can take a lot of forms, but the target rash is the classic one. They have some nonspecific symptoms like fever, fatigue, headache, and joint pain. And if it's untreated, it can lead to certain neurologic and particularly cardiac complications, it can lead to a cardiac arrhythmia. So that's why we're very interested in preventing this infection, and there's currently no human vaccine on the market. Why are we convinced that this could work through mRNA? Well, the program has actually already been derisked from an antigen perspective. So there was a licensed vaccine LYMErix that targeted the same antigen. That vaccine was later withdrawn from the market. It's actually the same antigen that we're targeting because it's a known mechanism of action. And it actually works a little bit differently than the viral vaccines we've been talking about. We are looking to induce primarily antibodies. And the protein that we target has the very creative name of outer surface protein A. The anti-OspA antibody that we produce, it's not actually protecting the individual person. It's protecting by transferring that antibody to the tick when the tick feeds. So it gets an antibody dose as it gets also your blood. And those antibodies can kill the Borrelia organism, the responsible bacteria for Lyme disease in the midgut, and that prevents transmission from human to host. So it's kind of a cool mechanism. And like I said, it's been established in a previously licensed vaccine. We have 2 candidates cleverly named after the dates or the years when there were various discoveries made about Lyme disease. The first recognizing Lyme disease in 1975 and then 1982, recognizing the pathogen that was causing it. So 1982 is actually a monovalent looking at serotype 1, which is basically the only serotype that is prevalent in the U.S. Europe is a bit more homogeneous and serotypes 1 through -- or heterogeneous excuse me, serotypes 1 through 7 actually proliferate there. And so a multivalent vaccine 1975 is targeted. So we had a Phase I study here, looking at both the multivalent and the monovalent so in red and in blue. It was a 3-dose study, given at 0, month 2, and month 6 and we're looking again in immune responses and the safety. And so here, you see the local and systemic solicited adverse events -- in this case, I told you we see some differences in antigens. We saw a few more grade 3s, particularly at the higher doses. And this is definitely a case where increased doses led to increased severity. There were no grade 4s that were reported. And most of the reactions are still grade 1 to 2 in severity. In terms of antibody, though, the vaccine really induced robust anti-OspA antibody responses. So you see here on the bottom, all the time points at which we measured, so people are negative at day 1. At day 29, they start to have some immune response, and it's really at that second dose that we start to see some increased day 57 to day 85. And then actually, there's pretty good maintenance of vaccine persistence, which we're kind of excited to see with a further boost, though that happens about at time 6 months. So we are seeing the robust antibody titers not only to serotype 1, which you see on the slide. Also to serotypes 2 to 7, and they did elicit dose-dependent responses. So there's a bit of a decision to make here as we look at balancing reactogenicity with immunogenicity. Both of the vaccines were generally well tolerated. They had an acceptable safety profile, and they elicited a robust dose-dependent anti-OspA responses. We've actually decided to go to Phase II in this program, and we're going to be evaluating not only lower dose levels because we think that -- even the lower dose levels, we're really inducing robust responses, but we're also going to be looking -- we're a platform company, constantly improving. We're going to look at applying some of those improvements that we've made in other programs to see if we can also improve that reactogenicity a bit. And with that, I'm now going to hand over to my colleague, Kyle Holen, who will talk to you about the oncology portfolio.

Thank you, Dr. Miller. And thank you all for being here in Cambridge and joining us here at our headquarters building, everyone online. Thank you for taking time out of your day to talk -- to hear about our pipeline updates. So we're going to have an overview of our oncology portfolio next. But I thought what we'd do before we talked about our oncology portfolio is just have a reminder of why this work is so important, but not from my words, from someone who's lived this experience. So maybe what we'll do then is quickly move on to the video. Can you show the video, please? [Presentation]

So I'm a medical oncologist by training. I spent 10 years taking care of patients with cancer. I can tell you that his experience is not atypical. Nausea, vomiting, severe fatigue, hair loss, diarrhea, infections, hospitalizations. These are all side effects of some of the most brutal therapies that we give people with disease. In fact, one of my patients came in and he unfortunately was not able to walk on his follow-up visit. He was on sorafenib, which causes very severe hand-foot syndrome. And I said to him, Why didn't you call me? Why don't you tell me this was happening. So I could have stopped your therapy, so you could have walked into clinic and he said, Doc, I wanted to survive. We have to come up with better therapies for people with cancer. Therapies that honestly can provide the same efficacy but not ask for the sacrifice on people's lives that they're currently going through with other therapy. And I think we're on to something with our current platform. So this is a snapshot of 4359 and 4157 or Intismeran of the safety readout from both of these programs. And I think a lot of us appropriately focused on efficacy of these products. But what I'd like to just do is take a minute and show you the safety of these products. So far across our entire portfolio of cancer antigen therapies and Intismeran, we have not reached a maximum tolerated dose. The incidence of Grade 3 side effects is vanishingly small. I can count on one hand how many times we've seen for monotherapy treatment, Grade 3 side effects from these therapies. And so I believe that this opens up a whole new window of opportunity for this type of treatment that hasn't been available for other treatments in the field. Areas where we can treat patients earlier in pre-cancerous lesions, areas where we can combine with other therapies that already have a fairly toxic profile where other treatments couldn't be added on. And this leads us to a whole variety of different settings where we can administer these therapies. So you can see here with Intismeran and other cancer antigen therapies, we've been able to explore adjuvant settings. We've been able to explore neoadjuvant settings. And for our cancer antigen therapy, we've even looked at early stage in precancerous settings for these type of therapies. With our T cell engager programs, we're looking at refractory metastatic settings and with our cell therapy enhancing and in vivo cell therapy, we're also looking at later-stage disease. So this allows us to create a portfolio that can have an effect across all different stages of cancer and multiple different types of cancers. This is a snapshot of our current oncology portfolio. What I'd like to emphasize is yes, I'm excited about the number of programs that we have in clinic -- but what's more important than the quantity is the quality. We firmly believe that every single one of these treatments can have a dramatic impact on patient lives. And as we'll discuss with you today, we'll start out with Intismeran, and Dr. Brown will come and talk to you about the latest developments with Intismeran. I will then come back and talk to you about our next most advanced therapy, 4359 which we're all very excited about. And then last but not least, we'll have our fearless leader for our research efforts, Dr. Lachlan come and talk to us about all the really exciting and amazing products that are coming into clinic either are in clinic today or will be coming in the clinic very soon thereafter. So with that, I'll hand things over to Dr. Brown, and she'll talk to you about INT.

So hi, everyone. It is nice to be speaking to you again about Intismeran, and I'm not really sure how I'm supposed to follow Kyle's performance and then Keith's video. So hopefully, what I show today is sort of our resounding belief of why we believe that Intismeran can make that type of impact for patients and a litany of patients at that. So as Kyle alluded to, we are leveraging the power of the mRNA technology and the platform to develop a precision immunotherapy program overall and we're anchoring this on 4 different segments with distinct modalities and distinct potential to impact patients across a wide continuum of cancer care. And the anchor and the lead of our programs are Intismeran autogene, which, as you saw on the slide, has a litany of clinical studies. So that is what we'll double down on today. And if you're not resonating with Intismeran, this one is the one that we called mRNA-4157, V940, INT, it's had a lot of different names and roles, but at its true form, it is an mRNA lipid encapsulated, individualized neoantigen therapy. So it is basically taking and starting with the patient and sequencing their tumor to identify the sort of most immunogenic specific tumor targets to train the immune system. It is one medicine for one patient, and it is the lead in personalized cancer therapy. The way this works is, again, what we're doing is taking a concatamerized mRNA lipid encapsulating them, delivering them back concatamer into patient through an IM administration. Once in the body, the mRNA enters into an antigen presenting cell, we use post-translational machinery to produce those neoantigens. Those neoantigens, which are basically cancer-specific mutations are presented on the cell surface to stimulate a repertoire of T cells, both CD4 and CD8s. That then become activated, but also trained to go out and seek other cancer cells or cancer cells and destroy them. And we think that this mechanism is synergistic with checkpoint inhibitors who just tend to take the brakes off the immune system. So we're essentially targeting cancer cells activating the T cells to go kill them and then hyper-activating them to keep them killing these cancer cells. So that mechanism was officially tested in our Phase II randomized study in high-risk adjuvant melanoma patients. This data was presented last year at ASCO from our median 3-year follow-up. And essentially, what we did is we took Intismeran in combination with pembrolizumab versus pembrolizumab standard of care in these high-risk melanoma patients who had their tumors completely resected and we follow them to see if the combination could improve recurrence events. So basically, patients' tumors from coming back after they had their tumors removed. One thing I want to point out is that this was a November 3, '23 data cut with that median of 3 years. We know in cancer that 5 years is usually the big landmark for clinical studies as a whole for durability. We are now in '25. So we add -- we're looking upon our 5-year data cut marginally. Now the one piece in here is that this data cutoff was on November 3 with the approximately 3 years. Because 5 years is such a landmark -- we are going to make sure we pass that landmark. And so the data cut here might be a little bit later this November 3, but we are definitely tracking towards having that. And we're very excited to see the durability because we are very excited to see the 3-year data. So this is a refresh from that 3-year cut. And basically, what you saw here is the primary endpoint, which was recurrence-free survival. The green line on top is essentially the combination arm. The yellow line is what happened with pembrolizumab standard of care. And what you're looking at is recurrence events or death. And so from just a global 3-year mark, we see a hazard ratio of 0.51 which means that we reduced the risk of recurrence or death globally for these patients by about 49%. And if we sort of anchor on that 2.5-year landmark, that means that 3 out of 4 high-risk adjuvant melanoma patients were disease-free at the time of this cut and that was 20% higher than what they would have received if they receive standard of care pembro. And it's important to note that, that delta of almost 20% is essentially what pembro showed against placebo, against nothing. So no other study has been able to showcase this type of effect above pembro standard of care in this type of setting which was very, very encouraging for us for this type of neoantigen therapy and the impact we could have on patients. And not only was this seen with recurrence-free survival, but this was also seen with the key secondary end point which is distant metastasis-free survival. And on this one, we see a 0.38 hazard ratio. So in oncology studies 0.38 is transformative. This is a huge treatment effect. You don't tend to see this, and it translates into almost 2/3 of patients not having distant disease. And distant disease is important because what that means is we're preventing patients from having significant surgeries, additional systemic therapies or death. Their risk is much higher when they have a distant event. And what you see here, again, at about 2.5 years is that almost 90% of patients on the study did not have this type of event. And that's compared to 68% that would have had it with just standard of care. So we're really maintaining that spread of 20% even in this profound endpoint. And this is actually on point with the mechanism of action that I just showed you because we believe that intismeran is supposed to train and activate the immune system to go out and clean up all these micro metastases and create long-term disease control. And so having this profound delta on DMFS is actually what we would have anticipated with the mechanism. And it's not just about preventing these recurrence events. It's also about preventing death, right? As we heard from Keith's story, when a patient is diagnosed with cancer, it is about survival. And so everyone really cares about survival. And while it's very early, we do not tend to see overall survival in these adjuvant studies, especially at 3 years. We did see an initial trend of protection for overall survival. Now the hazard ratio here is 0.42. But if you look at the spread here from 0.11 to 1.5 is big, and then that's just because there's just so few events. But what's encouraging about this OS trend is just how flat that green line is, and it was flat through that median 3-year follow-up. And it's one of the reasons we're so excited for what we're going to see with the 5-year data that's coming. And as Kyle alluded to when he started, we tend to focus on the efficacy picture. But for an adjuvant setting, that's curative intent. Patients want good quality of life, they want to be able to continue running and traveling and having that same life that they had before they were diagnosed with cancer. And so having a safe profile is amazingly important. And so not only did we see in this 3-year follow-up, this efficacy benefit, but we also saw that intismeran was generally tolerable where the majority of patients had low grade transient adverse events with fatigue being the most common. So not hair loss, not weight loss, not myalgia where they can't walk but fatigue. In addition, we did not see an increase in significant or serious adverse events when combined with chemo, nor did we see potentiation of the immune mediate adverse events which is what you see when you have IO-IO combinations. So it really is well tolerated as a whole and specifically for cancer patients having a clinical benefit and a well-tolerated safety profile is truly differentiating which gave us the confidence to move to this broad clinical trial program that we have now with intismeran with our partner, Merck. So what we have here is not only the Phase III adjuvant melanoma, which we'll talk about in a second. But we also have the 2 other registrational non-small cell lung studies that will hopefully benefit patients like Keith. And we have studies in renal cell carcinoma, muscle invasive bladder cancer, non-muscle invasive bladder cancer. Metastatic melanoma, metastatic non-small cell lung cancer and then our Phase I had expansion cohorts in PDAC and gastric. And what you see from this picture is that we're hoping to impact and learn across a litany of tumor types and a litany of settings. So we are really spanning with having solid foundations in the adjuvant environment. And then we're beginning to start exploring the bookends to metastatic disease and perioperative. And one of the reasons that we're actually comfortable looking at the metastatic space and the perioperative space, is basically what you heard from Jerh is we're optimizing our manufacturing. We're building our Marlboro facility. We're getting better at our turnaround times, and we have confidence that we will be able to deliver Intismeran to patients that need it most quickly. In addition, because of the safety profile that I just showed you, it means that we're also able to combine with agents beyond pembrolizumab. So in our metastatic non-small cell study, we're actually combining with pembro and chemo. Within our non-muscle invasive bladder cancer study, we're combining with BCG. In our Phase I study, we're combining with different chemotherapies for PDAC and gastric. So this program as a whole is really going to teach us a litany of things about intismeran tumor types, patient populations, combinations and really sets the foundation for us to have a profound impact. And what you can see based off all of this is that we have a series of learnings that are going to come '26 and well beyond. Now while everyone can pick their favorite study for a number of different reasons, our team is hyper obsessing about our adjuvant melanoma Phase III study. So this study was launched in 2023. And it looks largely like our Phase II study. So it is randomized in adjuvant high-risk resected melanoma patients. And it randomizes patients intismeran plus pembrolizumab versus pembrolizumab standard of care and has the recurrence-free survival endpoint, which is the appropriate clinical endpoint. The 2 major differences from our Phase II study is, first, we expanded the patient population down to Stage 2s, partly because pembrolizumab is indicated there and we have reason to believe that we can impact that wide range of high-risk adjuvant melanoma patients. And the other piece is, this is a stage III study, our Phase III study. So there's close to 1,100 patients that got enrolled. It was big, and it enrolled in record time. And I think one of the reasons that it enrolled in record time is that people are very excited for this Phase II results that I showed you. And so at this point, it has been fully enrolled for quite some time, and we are sitting in a very awkward phase right now where all of us are amazingly excited to see the RFS and to see the endpoint here. But this is an endpoint-driven study, which means that we also don't want these events to be coming in too fast because we want patients to benefit. So we are in this like, we really want to see it. We want to see this actual output to see how the Phase III is going to play out and if it looks like the Phase II but we have to be patient because we also want that treatment benefit. We want patients to benefit. So that is the dance we're doing, and we're just patiently waiting right now for those events to accrue. So on the clinical summary, what I basically told you to this point is that we have a manageable safety profile, we have a clinically significant improvement in recurrence-free survival. And distant metastasis-free survival. We have encouraging trends and overall survival. We have launched a series of clinical studies, 3 of which started this year with NMIBC metastatic melanoma and metastatic small cell lung cancer. And we have a line of sight for a lot of studies that are going to be coming '26 and beyond, starting with the 5-year median follow-up for adjuvant melanoma in but then each of those other studies are going to start reading out. Now behind the clinical learnings, we also, at Moderna, want to lead the science in this space. And so what I like about the intismeran program is that it sits at the precipice of computational, technological and biological innovation and understanding. And at its heart is our deterministic fixed algorithm, which selects the neoantigens for patients using their specific DNA sequences and rank orders, the ones that are going to be most immunogenic. And the thing about BICS is that it was perfect for the science at the time but as our clinical understanding evolves, as our understanding of cancer biology evolves and just as the field overall evolves, we have the opportunity to think or rethink about BICS as a whole. But that's going to take more science and more data. So built into our programs is translational endpoints or collections. So if you reimagine the clinical study schema that I showed you for the clinical side, this is the Phase II again. Nothing's really changed for how it looks outside of showcasing where we're collecting patients' blood to do some translational work. And so we had patients that gave blood right at baseline before they had any treatment. Patients then they have blood collected after they started pembrolizumab treatment. Then they had more blood collected after they started intismeran treatment. And then finally, at the very end of their treatment with pembro a year later as they had another sample collected and the data that we presented this year was 2 things. The first was to address the #1 question we keep getting from a lot of folks, which is do you really need an individualized therapy for neoantigens. Why don't you just off-the-shelf? Isn't that a lot easier? So that's the first question, and I'll show you that data. And the second was -- you've shown us immunogenicity, you've shown as you can mount the T cell response, but do those T cells matter, how does that population look as a whole? So that's the data that we presented this year. So the first part is addressing the idea about is a neoantigen therapy have to be individualized. So the first piece is in this Phase II study, the majority of patients did have an mRNA that spanned the full 34 neoantigens. So the BICS algorithm was working to be able to select the right neoantigens for the majority of these patients because melanoma is a high TMB tumor. They should have a lot of mutations to choose from. But that's not true for every patient. And so we're not forcing new antigens if they don't have them. And so that's why you have this sort of range. Now if we look at the bar graph, this is not patients, right? We don't have 3,401 patients in this study. But we do have 3,401 neoantigens that were completely unique to a patient which means that 99% of the neoantigens that we are picking that BICS is picking, it's clearly specific to each individual patient. And what that means is that no, we can't do an off-the-shelf approach for this. We have to tailor it to the specific person because the specific person's cancer is as unique as they are, and we have to generate a therapy that is going to treat them. The next part is we started looking at the idea of what do the T cells look like in these patients? Are they doing anything? Because core to the hypothesis is that we're selecting neoantigens, those neoantigens are training the T cells, the T cells are going to go and clean up the tumor cells. So we're using TCR sequencing to characterize the T-cell milieu of an individual patient and saying, does that change when a patient is treated with is intismeran. And we are able to do this because actually, your T cells are also unique and they each have their own unique fingerprint. So we can sequence them and see how they change over time. So this first slide is basically showing what the T cell clonal phenotypes of the population looks like at baseline for both the combination arm and the pembro arm and map that against if a patient had a recurrence event or not. And this is basically a negative slide that basically says, it doesn't matter what your T cells look like to start with that didn't impact the outcome at all. What that means is that the treatment effect matters because baseline your T cell and your immune systems is intact. So this is where we're going to squint but hopefully, you'll take my word for the story. And if you have interest, you can look at the SMR and the AACR picture. But if we're going to look for the waterfall plot first. So what we're doing here is now showing how the T cells change over time with intismeran treatment or pembrolizumab treatment. So the red is the top, and this isn't a combination. And what you see on the waterfall plot and what you're going to try to read is that 71% of patients that were treated with the combination had an expansion of their T cell population, which means that when they were treated with intismeran they actually had new T cells grow and adapt and their population actually shifted, which was different than pembro, which most of the pembro patients did not have that effect. And actually, if you look at not only the sort of broad population of T cells, but how many were new, which is what we would want with the neoantigen approach, right? How many new T cells are we forming. Those that were treated with intismeran actually had a higher proportion of new T cells, really, which is on point with the mechanism, and that was not true for the pembro, which, again, doesn't sort of fit pembro's mechanism. Now you can say, "Well, your patients were treated with pembro isn't it doing something?" And the answer is, no, actually, it's not here based off of the data because what you see is that screening the population -- the T cell populations are the same. When pembro starts, it doesn't really change. It's really once intismeran gets on board that we start seeing these novel clones. And not only are we seeing novel forms, but we're seeing clones that are only for those unique neoantigens that I told you about, not the shared ones. So if you put basically all of this together, saying our algorithm is picking neoantigens. Those neoantigens are stimulating T cell responses that are new and that are changing the milieu of the T cells throughout the body to be able to have long-lasting impact. Now the question is, what does that impact? Is it meaningful? Great, you showed me that you can expand populations that you want to? And the answer is yes, it's actually important. So what we see is these T cells are actually associated with recurrence-free survival. So what that means is it is important for intismeran to have an impact to have this clonal cell expansion because that means that this targeted -- tumor targeted patient-specific T cells are now going out and cleaning up whatever cancer cells are left in the patient's body to create disease control and prevent recurrence. And that this is true with a statistically meaningful impact. So all of this basically is to say that mechanistically, we have confidence in what we're doing with BICS and so not only are we seeing clinical impact and a good safety profile, but we're seeing that intismeran is doing what it's supposed to be. So with that for the translational summary, we're basically showcasing that the mechanism for what we hypothesized is actually playing out in clinic. But not only is it playing out in clinic, it's actually making the clinical data also makes sense and the fact that we are able to have this long-term disease control. And so the sort of next step here is to sort of tether this one step further and say, are all those T cells that I just showed you that are expanding do they actually track back to the neoantigens that we've selected and which ones and how can we optimize those neoantigens and how can we optimize BICS to create even better ones. And so that's why when Stephane and Steven were talking about the excitement we have for the oncology portfolio in '26 and beyond, it's really referencing the amount of scientific and clinical data that we will be coming with over the next couple of years not only to help our learning but to actually improve what we're doing as well. And so on that theme, the excitement for '26 and new data and the potential for the oncology portfolio, I'm going to hand this back over to Kyle to talk about our 4359 program.

So our most advanced program is 4359. We also have other therapies in clinic. That are part of our cancer antigen therapy portfolio, which include 4106 and then our Lynch syndrome program. So 4359 has 2 targets PD-L1 and IDO. These are targets that are well described and validated in oncology. I don't have to talk to you all much about PD-L1. However, I do want to take 12th to talk to you a little bit about IDO because we have some questions that we get about IDO and the importance in cancer. What I'll share with you is that we're not targeting IDO to change IDO functioning. Other programs that have targeted IDO have done so because they wanted to introduce small molecules that change the downstream pathways of IDO. What we're doing instead is we're targeting IDO through a mechanism where we can use it as a flag where T cells can be directed towards that express protein and use it to both have an effect directly on the cancer cell as well as have an effect on the overall immunosuppression that occurs in cancer. And so we can change the immune environment by targeting IDO and PD-L1 but we can also have a direct effect on the cancer cell. We presented our safety data at ESMO in 2024. And then most recently at ESMO in 2025, we released our first look at some efficacy data and an expansion cohort that we did in metastatic melanoma. These were patients that were checkpoint inhibitor refractory. So as you all know, checkpoint inhibitors are standard of care for patients with metastatic melanoma. All of these patients had received multiple prior checkpoint inhibitors. All these patients had metastatic melanoma. And they all received a combination of 4359 with pembrolizumab. So these are some design features of the study design as well as some demographics of the patients and some baseline characteristics. So we had 2 different dose levels that we assessed, 400-microgram and a 1,000 microgram dose level. As I mentioned, both of these were administered in combination with pembrolizumab, and we had overall 29 patients that were enrolled in the study. However, only 25 patients were eligible for assessing response to treatment. I talked to you a little bit already about the safety profile. But what I'm excited to share with you again is that grade 3 events were very uncommon, even in combination with pembrolizumab, we did not see any grade 3 events. And as Michelle described before, 1 of the concerns that many people have had, when you combine a treatment that has the potential to boost the immune system that you might get an increase in immune-related adverse events. But fortunately, for both the 4157 program and now consistently with the 4359 program, we have not seen an increase in immune-related adverse events. So this is the first snapshot of the tumor -- antitumor activity of 4359. When you look at the entire population, both the 400 and the 1,000 microgram patients we saw approximately 24% of those patients with a complete or partial response to therapy. The disease control rate was even higher and patients who had either stable disease or a partial or complete response, increased to approximately 60%, and you can see that on near the bottom of the slide. And we also have the number of prior therapies listed here, you can see that these patients all had many prior therapies and then progressed through these therapies. So there was a highly refractory population. What was more encouraging, however, was the data when we separated the patients by those who were PD-L1 positive by TPS or those who are PD-L1 negative. So on this waterfall plot on the bottom of the left side here, you can see patients who are PD-L1 positive represented in the red bars versus the ones that were PD-L1 negative in the blue bars. Every patient that had a partial or complete response was PD-L1 positive. So when you look at the overall population who are PD-L1 positive, you can see now the response rate goes from 24% up to 67%. And these responses were quite durable. And you can see that from the spider plot. So in the bottom right here, the spider plot has these patients separated similarly to the waterfall plot. For all the patients who are PD-L1 positive are represented in the red lines and the patients who are PD-L1 negative are represented in the blue lines. These lines represent changes in the size of their tumor over time. So when the lines come down, that means their tumors have shrunk. And the patients with tumor shrinkage had continued shrinkage of their tumor, out for 200, 300, 400, 500, almost 600 days. And just as a reminder, this therapy is administered 9x every 3 weeks. So the majority of these patients stopped therapy much earlier, but continue to have a robust and durable response to therapy. We've done some similar analyses compared to -- similar to what Michelle had described around T cell clonality as well as T cell-specific responses. I'll walk you through the data on the T-cell specific responses to PD-L1 and IDO. When you look at these slides, you can see the increase in the number of T cells with these either IDO directed T cells or PD-L1 directed T cells. The patients in purple were the partial response patients. Those patients in purple all had increases in their T cell-specific responses. Now what we can tell from this slide is there were other patients that did not respond to a T cell-specific responses. So our takeaway from this is it's probably necessary to have a T cell-specific response but not sufficient. And when you look at the novel expanded TCR clones, you can see for the patients that had a CR or a PR, there's quite dramatic increases in the novel TCR clones whereas the patients that had stable disease or progressive disease didn't have the same dramatic increases in the novel T cell clones. So this exciting preliminary data has led us to expand into multiple cohorts, this Phase I trial. So we now have an Arm 2a. Arm 2a is a trial where we have a combination of 4359 and pembrolizumab in frontline melanoma. Arm 2c is an arm where we are combining 4459 with ipilimumab and nivolumab. Arm 2b is an arm in patients who have second line and beyond melanoma all with PD-L1 positive disease. And this is a larger cohort to what we had originally described where all these patients will be selected for PD-L1 positive disease -- and then we're also moving into other tumor types. And so we have Arm IIb, where we're assessing the response in patients who have non-small cell lung cancer and have a TPS score greater than 50%. And this will also be 4359 administered in combination with pembrolizumab. Okay. So with that, I will move on to the next part of our presentation in oncology. Dr. Loughlin, who will talk to us about our early programs. Thank you.

Thank you so much. All right. So we are going to start by building on what Michelle and Kyle already shared with you and moving into the cancer antigen therapy portion of the pipeline. Starting with mRNA-4106. Now similar to 4359 but quite different from our individualized therapies, 4106 is designed to encode for antigens that are shared across many different patients. So this includes many different types of cancer and all of the patients within those types of cancer. So when we think about the design for this cancer antigen therapy, we specifically chose to include multiple antigens. So if you took a single patient, you would expect their tumor to express multiple antigens within this therapy. So 4106 is currently in a Phase I study in advanced solid tumors and we'll be excited to share data once we have those available. But I'm going to spend a little bit more time on the next program, which is Lynch syndrome because this actually takes our cancer antigen therapies even earlier in disease to the point that we are trying to intercept disease before it truly becomes cancer. So for those who are not familiar with Lynch syndrome, it's a heritable disease where these patients have mutations, so they have defects in the mechanisms your body naturally has to repair damage to your DNA. So it can be mutations in a couple of different proteins but basically, as your cells normally divide, occasionally, they make errors in your DNA. And your body has mechanisms that go in and correct that. But if you have Lynch Syndrome, that repair mechanism is deficient. So you start accumulating mutations over your entire lifetime. Now as it turns out, these patients have an incredibly high risk of developing certain types of cancer over their lifetime. So if you have one of these mutations, you can have about a 50% chance of developing colorectal cancer during your lifetime and you can have very early onset as well. So these patients go through considerable surveillance, colonoscopies every 1 to 3 years starting at really early ages to try to identify things like polyps in the colon and have those removed before they progress to cancer. Now what we've done in collaboration with the University of Oxford is identify these mutations, which happen to be shared between these patients. And so with this therapy, we're actually training those patients' immune systems to identify the cells with those mutations before that truly turns into a malignancy. So trying to intercept and train that immune system to remove those cells before these patients actually develop cancer. So we're excited with Oxford, we're looking forward to starting this study next year. Now I'm going to pivot pretty substantially in terms of the approach that we're taking to treating oncology. So we've talked a lot about cancer antigen therapies and the idea that we were training your immune system to recognize cancer cells and kill them. With T-cell engagers, it's a different approach. So we are actually looking to guide your immune system to those cancer cells. And we have 2 different types of T-cell engagers in our portfolio. The first focuses on proteins that are expressed on the surface of cancer cells. So I'm actually just going to start with the cartoon on the left-hand side here. So our lead program in this space is mRNA-2808 and it encodes for a T-cell engager that on the one side, binds CD3, so it binds to your T-cells. And on the other side, we've actually multiplexed this product. So there are 3 different targets that are present in multiple myeloma that we are able to go after with a single therapeutic. You can see them here, BCMA, FcRH5 and GPRC5D. Now we think this is really important in disease because if you have the ability to multiplex, you can really avoid antigen escape, which is a clear mechanism in cancer and you can account for the fact that not all tumor cells will express every single antigen. There is some heterogeneity in cancer and we can go after multiple targets at once. And this has actually played out in the field. So there are recombinant protein T-cell engagers already in use in multiple myeloma. If you take 2 of those and clinically combine them, it's already been demonstrated that you can improve response rates and that depth of response. In our lead therapeutic, we are able to multiplex 3 targets. Now I'm using the word multiplex intentionally, not the word combination because for us, this is 1 drug product from a regulatory perspective. We're advancing 1 product that can target 3 proteins. Additionally, as you move further into our T-cell engager pipeline, we can encode other signals that can help the T-cell engagers be even more efficacious. So we can include, for example, co-stimulatory signals that really help those T-cells be more activated and add another layer of specificity to that activation. Now what we're showing here on the right-hand side is data from nonhuman primates, where we were looking to see with 2808 depletion of a cell population that actually doesn't even express those targets very highly. These are healthy nonhuman primates but really demonstrated to us that we could truly deplete those populations and that we're seeing a nice durable effect until those populations are able to come back. And we're able to do this with both IV infusion as well as subcutaneous injection, which you can see here in the blue line. Now the second approach that I described for our T-cell engagers -- I apologize. So 2808 is actually already dosing patients in our Phase I study. And for that, of course, our primary endpoint will be safety. But given the patient population, we should have a pretty good read on pharmacodynamics or the impact on those pathogenic cell populations and the proteins that we secrete. And we, again, will be excited to share those data as soon as they're available. So the second approach that we're taking rather than going after proteins that are on the surface of cancer cells is actually focused on intracellular antigens, so proteins that are within the cancer cell. Now we're excited about being able to target those because it really opens up the target landscape for us. Proteins that are on the cell surface tend to be shared between cancer cells and healthy cells. There may be more on the cancer cells and that's why we target them. If we're able to go after intracellular proteins, it actually opens up a much larger pool of targets that we can pursue and these targets tend to be very specific to tumor cells. So as Kyle and Michelle talked about, that safety and tolerability profile is really important for patients and we think this will improve that as well. We are still able to multiplex and go after multiple targets with this approach or to encode some of the supporting proteins like co-stimulatory factors. And so we're pursuing this in partnership with Immatics. Okay. We're going to transition to yet another part of our pipeline. We're going to talk about our approaches in cell therapy. And I'm actually going to start by describing something that Moderna does not do, which is ex vivo cell therapy. So our first approach in the cell therapy space is actually looking to enhance the performance of a partner's ex vivo cell therapy. So for those who are familiar with ex vivo cell therapy, you might have heard about CAR-T and in certain blood cancers, CAR-T therapy can be absolutely transformative. Now in solid tumors, these ex vivo cell therapies haven't delivered quite that same level of impact. And so we're looking to truly improve those outcomes with a focus in that space. So how do we do that? You first start with your typical ex vivo cell therapy. So you need to take the immune cells out of your patient. You need to engineer those immune cells. You need to try to remove the remaining immune cells in the patient to try to conceptually make more space for those engineered T-cells to engraft as you infuse them back into the patient. Now after the engineered cells are back in the patient, we administer mRNA-4203. It's an intramuscular injection and we have designed it specifically to encode for the antigen that is recognized by that engineered T-cell therapy. So mRNA-4203 is specifically designed for anzu-cel but it is not specifically designed for any individual patient. It will work for all of those patients. And you're actually taking those engineered T-cells after they're in the body and boosting them. So you're showing them the antigen that they are engineered to recognize. So they will get activated, they can proliferate. And we think this has the ability to enhance the performance of those engineered T-cells because it has been shown clinically that when you have these T-cells that are in a good immune state and that they are able to persist for longer in the body that those tend to be correlated with better clinical outcomes. And we are in an active Phase I study in collaboration with Immatics for this combination and this is a true combination of mRNA-4203 and anzu-cel. So our second approach in the cell therapy space is quite distinct from ex vivo CAR-T. It's actually in vivo CAR-T. So as I said, we do not do ex vivo cell therapy. Our approach here is to actually engineer those T-cells inside the body in the first place. So for this approach, we actually use a lipid nanoparticle that is specifically targeted to T-cells. So it's specifically going to T-cells and it encodes for that same CAR that you might have engineered outside of the body. But we can do all of this inside of the body. And importantly, this means you do not need to take the cells out of the patient. They do not need to have these tough conditioning regimens that take out the rest of their immune cells. You don't have this large, individualized manufacturing component. All you do is infuse LNPs. So we see this as having substantial advantages there. We can also rely on some of the differentiation that I described for T-cell engagers. So for example, we can multiplex. If you wanted to encode a CAR for multiple targets, you could do that with this platform. If you wanted to encode other proteins that would help these T-cells stay activated and go into that tumor microenvironment and be very efficacious, you can also do that. And we see 2 application spaces for the in vivo CAR-T style approaches. The first is actually in autoimmune disease. So there's been quite a bit of really promising clinical data of late that if you go into autoimmune diseases like lupus and you actually use CAR-T to eliminate and deplete those B cells, you can actually help those patients reset their immune system. So they can go into remission for a remarkably long time. So we think that's certainly an application for in vivo CAR-T, where the in vivo profile as well from a safety perspective will be really important to those autoimmune patients. We also think in vivo CAR-T will be highly relevant as you go into oncology. As I mentioned, CAR-T in general has been incredibly transformative in the blood cancers and we believe this will help us go into solid tumors as well. And our third approach in cell therapy actually focuses on a different cell type. So where CAR-T is very focused on T-cells, CAR-M is actually much more focused on a different cell type, myeloid cells. But the approach is still similar. So we infuse patients and using LNPs, we transfect different types of myeloid cells and we're encoding a CAR. Now those cells will move around the body and traffic. And when they get to the tumor, that CAR will recognize the antigen. Those myeloid cells can then actually ingest some of the tumor cells. So they're killing some of the tumor cells. They will secrete other proteins that will help other cells in that tumor microenvironment to be able to kill cancer cells. And then having actually engulf and ingested a tumor cell already, the cells can present multiple antigens from that tumor cell. So Michelle talked about how you get that broad T-cell response even with INT. This is a similar concept where even though we only encoded a CAR that recognizes one antigen, by the time you get to this part in your mechanism of action, you're able to really expand that response to lots of different antigens that are present within the cancer. So I know we moved quickly but we did want to have a chance to share this part of the oncology portfolio with you and give you a sense of the diverse set of therapeutic approaches that we are able to take because we can leverage so many different aspects of our platform technology. So with that, I believe I'm handing it to Dr. Rita Das. Oh, no, we're going to take a break. Lavina said we're going to take a break and bring lunch in.

Yes. So if I can invite everyone to please grab a plate of lunch and a beverage of your choice and bring it back in here and we will resume the rest of the meeting. Thank you. [Break]

All right. Hi, everyone. My name is -- I'm going to bring us back from break. My name is Rita Das and I'm the Clinical Development Head for Respiratory and Rare Diseases. And it really gives me great pleasure to focus on our rare disease portfolio today because as a pediatrician, I've taken care of these children with inborn errors of metabolism and seen the impact on their lives and also the underlying progression, neurologic and otherwise. And I'm really excited by Moderna's commitment in this space and I'm very excited that we're getting closer and closer to bringing these therapies forward. So first, I'll talk about propionic acidemia, which is where we're the most advanced. And propionic acidemia is a rare metabolic disorder, primarily diagnosed in infancy that causes a huge amount of morbidity and mortality. It's very rare, ranging from 0.29 to 4.24 per 100,000 newborns. So it's actually formerly in the ultra-rare category. And it's caused by pathogenic variants in the PCC enzyme. There's 2 subunits, PCCA and PCCB that stand in the way of the enzyme in the metabolism of proteins. And so when that metabolism is not going properly, you build up these toxic metabolites. And so because this is so pervasive, PA is a multisystemic disease with not only these metabolic decompensation events, which can be life-threatening but there's underlying neurologic, cardiac, endocrine and immunologic manifestations. Now there are no approved therapies for propionic acidemia. And so the management involves severe dietary protein restriction. And as the disease progresses, the patients often progress to needing liver transplantation. Now here's a bit more on the PA biology, which really shows why mRNA therapy is particularly well suited to succeed in this space. And so as I mentioned before, there's 2 components of that enzyme, PCCA and PCCB that come together and allow people to metabolize particularly proteins. And when that enzyme is not functioning, you build up these toxic metabolites that are damaging to the brain and other organs. And for mRNA-3927, what we're able to do is encode the functioning both PCCA and PCCB, package it in a lipid nanoparticle and deliver it IV but target it towards the liver, where then the patients are able to make these enzymes and correct their inborn error of metabolism. And so first, I'll talk about our Phase I study, which is called Paramount. It's a global study that enrolled in multiple countries. And the analysis from this study was presented in the ICIEM conference in Kyoto, Japan earlier this year. The primary endpoints for the study were safety and tolerability and a key exploratory endpoint was the reduction in metabolic decompensation events and I'll tell you a little bit more about that after. The study design was a dose escalation design. So we started from a dose of 0.3 milligrams per kilogram delivered every 3 weeks and we progressively increased that dose to 0.9 milligrams per kilogram delivered every 2 weeks. Now the key inclusion criteria for this study were, participants had to be greater than 1 year and had to have a confirmation of the PA genetic defect. The key exclusion criteria were if the patients were in grade 3 or 4 heart failure, which unfortunately happens in PA or either had a planned or a history of liver transplantation. So now here are the demographics and baseline characteristics of the patients. So 20 participants were enrolled in this study, 18 completed treatment and 17 participants entered the open-label extension study and 10 are continuing to receive treatment even today. The mean age was 11 years and the range in age was from 1 to 26 years. Now mRNA-3927 was well tolerated and had a manageable safety profile. In this study, we've administered almost 1,000 doses of mRNA-3927. And we've seen no drug -- dose-limiting toxicities. We've seen some serious treatment-emergent adverse events as would be expected in this population of chronically ill children and young adults but the adverse events that are related to treatment have been much fewer. Here's a little more detail on the adverse events that are -- that have been emerging. And these are fairly expected in this age group and those that are related are mild to moderate. We also saw very few infusion reactions and those were managed with conservative therapy. So just an overall summary for our PA program to date is that we've enrolled, sorry, 20 participants in Part 1, 13 participants have been dosed for over a year. There's been 43.6 years of cumulative patient experience of the study drug. The longest treatment was 3.1 years and the median duration of treatment was 1.45 years. [Audio Gap] Here's a little bit more on these metabolic decompensation events. These metabolic decompensation events have occurred in both PA and MMA, which I'll talk about next. And they're usually how these patients present first to medical care. They're either screened -- they're either identified on their newborn screen or they present with these metabolic decompensation events. And they're a major contributor to both morbidity and mortality and also to these long-term irreversible sequelae such as brain and cardiac damage. Now we had to discuss with the agency and agree upon a definition for these metabolic decompensation events, which we've done now. And so the definition that we've come up with is that the signs and symptoms include vomiting, anorexia, lethargy and seizures. There's also these observations of metabolic acidosis, which is a buildup of acid in the blood or high ammonia. Also, there's often a need for acute medical care, emergency room visits or hospitalizations. And so this is the scope of the definition that we've agreed upon with regulators. And this is what's really, really exciting about the data to date. The red dots represent the spectrum of metabolic decompensation events in the study. And on the y-axis, you see the different doses. And on the x-axis, it's over time. So pretreatment is before that black vertical line and posttreatment is after that black vertical line. And it's individual within patient comparisons. And you can see pretreatment, you see a lot of these red dots. So these patients are having multiple metabolic decompensation events every year. And as you get past the vertical line, you see these events really decrease in frequency. And as you go higher in dose, the 0.6 milligrams per kilogram and the 0.9 milligrams per kilogram, you see these events virtually go away. And so when you do the statistical analysis of this, you see that across all doses, the mRNA-3927 is related -- is associated with a 76% relative risk reduction in metabolic decompensation events. And when you look at just the doses above 0.6 milligrams per kilogram, that comes up to an 83% relative risk reduction that's statistically significant. And so this is what makes us really excited about our pivotal study, which has completed enrollment. So in summary, mRNA-3927 is well tolerated at all the doses administered with no dose-limiting toxicity. All the IRRs were grade 3 or lower and resolved with conservative management. mRNA-3927 treatment continued to demonstrate sustained reductions in these metabolic decompensation events with the highest benefit seen in those patients dosed at 0.6 milligrams per kilogram and higher. And these findings support the clinical development of mRNA-3927 at that dose, 0.6 milligrams per kilogram for the first treatment for patients with propionic acidemia. The registrational study is ongoing. This is Part 2 of the study that I already presented. Target enrollment has been reached and we're really looking forward to seeing the results next year. Also, since PA often -- since some of the neurologic decompensation happens in infants, we're also doing a dose-finding study in infants to bring the greatest benefit of mRNA-3927 across the spectrum of age in these patients. So now moving on to MMA, which is a related organic acidemia. Like PA, MMA onset occurs very early in life and is associated with these MDEs and then this underlying chronic toxicity. The MMA defect occurs a little bit farther down the pathway than PA but the enzyme that's responsible is the MUT enzyme. And this -- and the downstream sequelae are very similar. It results in a buildup of these toxic metabolites from proteins and fatty acids because of an inability to metabolize them. Again, protein restriction is the mainstay of therapy because there's really nothing else curative available. Some patients have levocarnitine supplementation and many of these patients progress to liver or kidney transplantation. So the MMA, the 37 -- the mRNA -- the MMA therapy 3705 encodes the MUT enzyme. So it replaces the MUT enzyme. It's packaged in the LNP just like the PA. And the LNP is specifically traffic to the liver where replacement of the enzyme restores the metabolism and the buildup of these toxic metabolites. So I'm going to present to you the Phase I/II study. That's a dose-finding study for MMA. This was a first-in-human study and that was -- it was enrolled globally as well. And it was also a dose escalation study. So we started with a dose of 0.1 milligrams per kilogram administered every 2 weeks -- every 3 weeks, sorry. And then we progressed to 1.2 milligrams per kilogram administered every 2 weeks. The key inclusion criteria were, again, MMA that was genetically confirmed to be due to a MUT deficiency. The age was greater than 1. And then the key exclusion criteria were very similar. Children with the background chronic disease that was too advanced were excluded and children who had a history of organ transplantation also were excluded. And so here are the demographics of the cohort. 18 participants were enrolled across 6 countries worldwide. The mean age was 7 and the age ranged from 2 to 39. Now there's 2 different phenotypes of the -- major phenotypes of the MUT enzyme, MUT 0 where the enzyme is completely absent and MUT minus where the enzyme is not completely absent but has a decreased function. And so here's the safety profile. The median duration of treatment was 99.6 weeks, so just above 2 years. All 18 participants finished their dosing in that base dose-finding study and continued into the extension study. Just above 860 doses were administered. The total patient years of exposure was 36.17. The safety profile was overall well tolerated. The infusion reactions also were easily managed and there was no dose-limiting toxicity. So again, 18 participants have been dosed. The longest treatment duration is 2.3 years. The median duration is about 2 years. It was well tolerated and all the participants are continuing in the extension study. Now here -- so for MMA, there is a plasma biomarker that we are hoping to use for our Phase III study. And we see that -- and that's the plasma methylmalonic acid level. And we see here that there is a greater than 50% decrease from baseline for participants who were treated with the mRNA-3705 at doses greater than 0.4 milligrams per kilogram every 2 weeks. And the plasma biomarker is more evident on this slide. And you can see here the greatest evidence of the plasma biomarker decrease is seen in those MUT 0 participants who are on the bottom. And as you can see, almost all of the MUT 0 participants, all the different doses are in the different colors on the right-hand side. The plasma biomarker, as you go up in dose, you see a very consistent decrease in the plasma MMA biomarker. And so it's just the lowest doses that have the higher residual. This is a little bit less clear in the MUT minus phenotype. Now again, the thing that we're most excited about in MMA as well is this reduction in these MDE events. This chart is set up the same way as the PA chart. The doses are increasing on the left-hand side. Time is on the x-axis. And you can see pretreatment and posttreatment. And pretreatment is in the red circles, or the MDEs are in the red circles. And as you can see, pretreatment, these children are having -- these children and young adults are having quite a few metabolic decompensation events. And as you go to posttreatment, you see a 91% relative reduction in the MDE events and a 75% relative reduction in MMA-related hospitalizations. And so we're very excited about our MMA preliminary dose-finding results. And so in summary, for MMA, mRNA-3705 was well tolerated in participants with the MUT-deficient MMA in this study. There were no dose-limiting toxicities and no treatment-emergent adverse events that led the patients to discontinue the study. We saw reductions in disease-related biomarkers that was the plasma MMA levels that indicated improved metabolism. And we also saw the reduction in the MDE events as well as the MMA-related hospitalizations at all doses greater than 0.4 milligrams per kilogram. MMA is ready for the pivotal Phase III and we aim to begin that in 2026. So with that, I'm going to hand it back to Stephane.

Thank you, Rita and thank you to everybody who presented this morning. As I said this morning, I'm going to be quick, 2 slides. The first one is, as you heard me this morning, our near-term strategy is clear, is drive sales growth through the respiratory seasonal vaccine franchise, drive profitability, drive cash to invest in oncology and rare. You've seen from the presentations that the oncology portfolio is really exciting and extremely differentiated from what you can see from other companies. So we're really eager to see what this does for patients and how we can help patients. And as Jamey [ closed ], if you think about what we're trying to do is drive the top line, drive gross margin improvement, evolve R&D investments to diversify further away into oncology, reduction of project cash cost through every year in the process and really drive the company back into profitability until '28 and from there, deliver a lot of patients impacts. So with this, I'm going to ask my colleagues to please join me and we'll be delighted to take your questions.

We will take questions in the room and as well as online. [indiscernible] yourself and tell us your question.

Yes, I have to say I never thought I'd have so many questions about manufacturing. But in the interest of time, I'm going to kind of skip to some of the other stuff that I jotted down throughout the morning. So if I may start, just with respect to the 1083 combo vaccine, what is the FDA waiting on? What do you need to hear from them? And is there a potential risk that you might have to do another study?

Yes. So look, our efficacy data are relatively new. The FDA has not had the opportunity to review that data in the very in-depth way that they do as part of their review. So I think what they really want is that we submit this flu file and really get to check under the hood. And then we certainly will have additional discussions.

This is [indiscernible] here for Geoff Meacham at Citi. So with the new debt facility, you guys mentioned that it's going to further bolster the balance sheet and open up optionalities. And so with that, what future opportunities could that provide? And how might that share -- shape your view on which pipeline programs you might bring back online and/or partner out?

Yes. Thanks for the question. So I would say we don't know right now. We're prepared for both upside and downside. I think that's what this does for us. Right now, we plan to park the cash on our balance sheet and therefore, bear a small spread from an interest rate perspective. And then we'll see how the next few years unfold, whether that's pipeline increase -- on the opportunity side, whether that's pipeline increase, whether that's share buyback or BD, we'll see how that goes on the opportunity side. We'll also understand what's happening with our revenue line. We need to execute our base plan first. So our base plan is to breakeven by 2028 and that is what we are going to monitor to, to make sure that we do that. And so this facility provides flexibility both for the opportunity side. But if there is something that happens, we're not pointing to anything. We believe in our base plan. But should something happen to the downside, it also provides flexibility on that side as well.

Yes. I mean if I could just add, I think as you asked the question about where we'd focus in the pipeline, we do have a large number of products to launch in the vaccine space, as we talked about. But I think as Jackie and Rose and the team highlighted, there are some exciting programs either in the mid-stage development in vaccines, our EBV vaccine, Lyme and others and our early oncology pipeline that we're equally excited by. We're going to wait until we're ahead of plan on cost, which we've been lately and showing we can grow that top line. But having the flexibility and the facility to be able to invest when we have very high confidence that we're ahead on that breakeven target is going to give us an opportunity to accelerate that mid-stage pipeline.

Tyler Van Buren, TD Cowen. I wanted to start on the financial side. So can you elaborate on the assumptions behind the 10% revenue growth year-over-year in '26? I think that's a little bit above consensus. The slide says it doesn't include flu, even though I think consensus does. So just curious to know if that growth is coming primarily from the partnerships piece with the U.K., Canada and Australia? And do you have confidence in that because you're already deep in those discussions? And is that assuming stable COVID revenues or decline? And then the follow-up to that would just be with respect to your year-end cash balance exercise through '28. That clearly has some sort of revenue assumptions through '28, for '27 and '28. So how are you thinking about revenue in those years as well in that exercise?

Yes. I'll start and then feel free to add, Stephen. So as it pertains to the up to 10% growth in 2026, there was a couple of questions in there. But the short answer is, yes, it primarily is driven by those 3 facilities, which is what Stephen tried to lay out today. We're not only in advanced discussions. We actually have it contracted and built. So these facilities are built. The contracts are already in place. So -- and we're already executing on the contract for all 3 of them but there is some revenue as it pertains to Canada in our assumption this year and a little bit in Australia. If you remember, for the U.K. earlier in the year, we pushed out over $200 million in revenue. So that will take place next year. So if you just think about 10% growth by itself, if we're $1.6 billion to $2 billion or $1.8 billion at the midpoint, $200 million is 10% growth. So that alone is sufficient. Now we will have a full annualized impact for all 3 contracts. So therefore, it will be a primary driver of growth next year. So that's #1. Next spike share, as Stephen laid out earlier as well, is the other primary driver. You asked if COVID is flat, I think as it pertains to the U.S. We can basically offset further decline actually. So with the -- what we believe will happen outside the United States with the 3 contracts I just talked about and other countries but these are 3 primary contracts. We actually believe we can offset and still grow up to 10% with the U.S. decline. Now we're hoping it doesn't. And we're hoping next spike will increase our share and value. But nonetheless, we still feel confident even with a little bit of decline. As it pertains to walking forward the cash balance, yes, obviously, that has some revenue assumptions in there. So I said that we would invest $2 billion in the year 2026 and I said $4.2 billion of a cash cost target. We're $1.6 billion to $2 billion on a revenue side, so take $1.8 billion and if you're at the midpoint. And if we grow up to 10%, that's $2 billion. So that's technically a $2.2 billion net cash investment but I'm kind of rounding here. And we'll see where this year lands. So if this year lands on the high side, then certainly $2.2 billion could be in play if we're at $2 billion plus 10%. If we're on the low end, then we'll have to relook at it and understand what will happen. But I still feel good about the overall $2 billion. As you look out a year, we said that we would invest net $1 billion to $1.5 billion with a cash cost of $3.5 billion to $3.9 billion. So on the low end, a $1 billion cash burn would imply $2.5 billion of revenue. So if we had $3.5 billion of cash cost and $2.5 billion of revenue, that's obviously $1 billion of cash burn. But we're not trying to be specific about that. We're not guiding that number. We will continue to ebb and flow. 2 years from now is a long ways away. In 2 years, we took out $4.5 billion of cost. So what can happen in 2 years? There's a lot that could happen in the next few years, which is why we're not guiding a revenue line but it gives you some semblance of understanding. We do expect revenue growth, not only in '26 but also in '27 and '28 and Stephen laid out the drivers earlier.

That's great. Remarkably detailed as always, Jamey, appreciate it. I have to ask about INT before I pass the mic. Watching the Marlborough facility video and having Jeff walk us through the optimization, it's clear that you all are not only very excited but have invested significantly in the future of intismeran and INT in general. So can you just elaborate on, one, your confidence in the Phase III readout next year and the powering of the primary endpoint? Two, if you need to show an OS trend for approval? And three, how confident you are that it will come next year based upon what you're seeing with the events and where you're at with the events?

Maybe I'll take a first stab at this and please ask my colleagues to chime in. So this is probably one of the most derisked Phase III programs that happens in oncology because we have a randomized Phase II trial that showed an incredible hazard ratio of 0.51 in RFS at our 3-year data point. So given that derisking, I'm very confident in the Phase III program being successful. But this is a blinded study. So I can't tell you what the results are until we finally get to our first interim and then can unblind and check. In terms of the powering, so we've made some fairly conservative assumptions on powering that study. And those conservative assumptions are much higher than a hazard ratio of 0.51. And I think that gives us even more confidence that we'll hit the target that we're hoping to achieve. Lastly, we currently are on track with our events to what we had discussed in previous guidance, which is third quarter of next year is when we hope this matures and we can do our first analysis. However, event rates are sometimes unpredictable and we really need to wait until all those events can happen before we do our first analysis. And so depending on the continued trend on those events, we'll have a better idea of when that first analysis will take place. But right now, we expect it to be in that third quarter next year time frame.

I would just add a couple of things. So it could come sooner than that. And in fact, we're -- we really don't know. It's an event-driven trial. And so throughout 2026, we'll be looking for it but we do expect it to happen in 2026. I would highlight that there are a couple of other either fully enrolled or largely enrolled studies. We highlighted some of them today, Phase II studies IIb randomized studies that actually also could read out in 2026 and provide strong confirmation of what's possible with INT. And so the way we look at 2026 is there will be multiple readouts of mid- to late-stage studies, obviously, the first being the Phase -- well, not the first necessarily chronologically but the one most anticipated being INT for melanoma but a lot of data coming in, in very short order. And so we're looking forward to 2026. All of them event-driven. So we'll wait and see.

And [indiscernible] the last piece is also, this is not my -- our investment, it's investments with Merck as well. As you remember, Merck is investing half of all the investments in manufacturing. Marlborough, we are running it but Merck is financing half of it. And of course, for the studies and all the medical commercial readiness and so on. So it's actually a huge commitment from a company who I think does a few things about immunology.

This is [indiscernible] for Cory from Evercore. Just on the topic of intismeran in adjuvant melanoma. How do you view it in the context of the emerging data and potential use of pembro in an earlier neoadjuvant setting?

Yes, it's a great question. So we've looked very closely at how much neoadjuvant use is happening, both for pembrolizumab and for nivo and ipi. And right now, we haven't seen a dramatic trend in those neoadjuvant rates. So we don't believe that this is going to be a major factor. Those are also not approved uses of pembro or nivo or ipi. They're not labeled. And there's no plan from these companies that I'm aware of that they're planning on changing the label. So if intismeran is a positive study, it will be labeled as such and it will be something that we can discuss with prescribers about the impact that this can have on their patients. So I'm confident that we'll be able to make sure that this is changing the treatment landscape based on positive Phase III data and based on a label that explains that.

Yes. This is Huidong Wang on behalf of Gena Wang from Barclays. We have 2 questions. First one, regarding ongoing IT litigation with Arbutus, does it apply to the next-generation spike? And for Section 1498 defense, what's your strategy to defense for the government rather than for the American people? Second question, it is very exciting to see these new discovery assets. Given the cash guidance of the breakeven in 2028, what's the priority list for all of those assets? So the question is for assets. Since we need individualized new antigen, what's the process time now and which steps could be improved?

So let me start with the first question. As I said during the Q3 call because we got similar questions, we believe in our IP and we're going to be defending it actively. I won't comment further given there's a litigation coming soon.

I'll take some of the pipeline question. Look, I think we'll be data-directed. We are very excited to move forward with the mid-stage vaccines, with the early-stage oncology programs as we get clinical data over the next year or 2. It will ultimately depend upon where we see the most compelling opportunities in the data from those ongoing clinical trials. Hard to speculate, but a large number of opportunities to invest as we start to continue to grow in '27 and beyond. And there was a turnaround time. It was -- a question was sort of how are we doing on delivery and turnaround time for INT and any opportunities to further improve that as we go forward from a manufacturing...

From a manufacturing point of view, we're doing really well. I mean, compared to where we were 12 months ago versus now, we've improved our manufacturing turnaround time by 50%, and we have plans to even continue to do that more. So I'm looking at the whole needle to needle, we have plans to continue to drive that forward as well. So not only will that give us further capability to be there for the patients and their need, but allows us also simultaneously to drive efficiency.

Just to give you a sense, across our clinical trials, there's 1,000-plus patients. You saw some of that data. And we are already ahead of our target turnaround time for commercial launch. And so we're quite confident that we're already performing at a level that we need to. We'll always look to do better. But it's actually even better than we saw in the Phase II study. And so we're pretty confident we can achieve our target product profile.

And this is across multiple countries around the world. So over 40 countries, we've been able to continue to achieve these really robust turnaround times.

Myles Minter from William Blair. Jamey, I think we were here last year and you put a $6 billion cash cost basis number up on 2028. Obviously, I haven't seen the 2028 number yet, but I assume that's going to be closer to $3 billion than it is to $6 billion. What's changed in 12 months that gives you the confidence there? And how much of it is just efficiency versus a need to get that low because revenues are maybe not projected to grow as much as you thought 12 months ago?

It's a good memory, yes. So we did say that we would break even in 2028 by $6 billion -- at $6 billion. I think it's probably a bit of both, frankly. We continue to say that it is both prioritizing the pipeline as well as driving efficiency, to which our teams have done it faster and to a much greater level than what we could have ever anticipated. So I think it's a little bit of both, frankly, that instead of being at $6 billion, I don't know if we'll be at $3 billion, but we'll certainly be at least $3.5 billion to $3.9 billion. So we have been able to take it down, and it is a mix of both, but I've been super encouraged by the entire team, how everybody has done it across every single organ function inside the company. And we continue to see greater opportunity, like I said earlier, more than we thought we could do and faster than what we thought we could do. So that's a big driver of it. But we definitely had to prioritize the pipeline as well. But everything that you've seen across all the products, you've heard our story. The candidates that we think will read out over the coming years that we believe will be launched in those respective years, '26, '27, '28 have always been prioritized and will continue to be prioritized. We made a decision on a lot of the Phase II trials, which is what Stephen was just referring to in the answer to the question, that we will not be able to take those to Phase III until we actually see that we can break even and can afford those Phase III trials. So that was some of the prioritization that we had to do as well.

This is Elizabeth Webster from Salveen Richter's team at Goldman Sachs. And we wanted to ask about the opportunity for RSV to come into play in the growth story in a bigger way and could that be an upside lever? And then framing what you would like to see at the interim for the norovirus Phase III next year?

Okay, I'll take the first step. We do hope RSV becomes a growth driver as well. It is built in to some of those strategic partnerships. And so when we highlight U.K., Canada and Australia, I'll remind you, those are across our entire respiratory portfolio. So not just COVID and RSV and flu and the combination. All of those are built in. And so it is an important driver internationally to expand. And as I highlighted, as we move into Europe, we do expect it to be a contributor as well. In the United States, it has been -- there was a very rapid launch in that first year, and there has been a bit of a waiting game for the revaccination year for our opportunity to present itself. We did show up a year after that first wave of vaccination. We have had some wins. We won a VA contract, we're slowly and steadily adding to our share, and we're getting ready for that revaccination, but that determination will be made by public health, not by us. So as we expand internationally, we do expect RSV to be a contributor.

And then in terms of norovirus and what we're hoping to see at the interim analysis, I think it's pretty typical of what we want to see. We want enough cases to be adequately powered to give us a potential shot on goal in terms of vaccine efficacy. We also want to conduct a futility analysis. So we're investing in a cohort here to capture a certain number of cases. And we want to understand from our DSMB without unblinding data, but just are we in the right direction or not. So we anticipate that later next year.

I'm Lili Nsongo from Leerink Partners on behalf of Mani Foroohar. I wanted to touch on the projection in terms of cost reduction, specifically as it relates to R&D regarding the vaccine franchise specifically. So you mentioned that you expect a reduction given the completion of the Phase III studies. I wanted to understand, can you give us a little color in terms of what type of post-marketing studies are baked into your assumptions as it relates to the 2 COVID programs as well as the flu program and the combo vaccine.

Yes. So thanks for the question. So we actually did bake into those assumptions the need to do some post-marketing work. So our assumption is that we will be starting actually imminently, the post-marketing commitments that we have in COVID. One of them is actually already well underway. The second one is starting soon. And then we have planned that in flu, we may need to do some post-marketing work, certainly on the safety side, potentially on the effectiveness side as well. But I think the flu efficacy data is really what we were waiting for. It is controlled against flu vaccine that's licensed in the U.S. So we're really confident in the strength of those data.

Maybe just a follow up for the COVID component does it include post-marketing studies for the below 65H?

It does, actually. So we have taken FDA guidance, had some back and forth on what they need to see from that study design. And like I say, one of them, the immunogenicity one is ongoing, and we're looking to start the effectiveness one soon.

And I think Jamey alluded to that, there is a little bit of an artificial elevation of that number compared to what we expect the maintenance to be in '26 and a little bit into '27, related to the new post-marketing commitments that the U.S. FDA asked for. Outside of the U.S. and the rest of the world, we've already really transitioned to that more standard post-marketing number.

Great. I'll try to squeeze in two, if I may. So first, I really enjoyed the redosing data on mRESV. I found that really compelling. Where are the regulators currently in terms of considering, recommending redosing for adult for the currently approved population? And then I have another quick kind of high-level question, if I may.

Yes. I think maybe I'd say it's probably not a regulatory decision on when revaccination happens. And again, this is revaccination. And so the decision on when we see waning of efficacy from a respiratory vaccine and when that public health benefit is sufficient to justify paying for an additional dose in most countries will be made by NITAGs, not by regulators. Ultimately, in the U.S., that's the CDCACIP. What you're starting to see in some of the real-world effectiveness data is, and this has come out for all 3 vaccines, is there's pretty substantial waning by year 2. And by year 3, we'll look to more data, but we do expect it to be down. If you have high risk factors, immune compromise of some form, it's actually happening even really after 1 year. And there's a rationale to be made to put forward for protecting those at higher risk even more frequently than, let's say, every 3 to 5 years. So I'd look for that to be the first place you'll see movement and probably you'll see it through the recommending bodies first, not regulators.

Yes, that makes a lot of sense. And then kind of a high-level question. So as we think about individualized neoantigen therapies versus shared antigen therapies, what do you really see as the future playing out? Like the immunogenicity data that you shared was pretty compelling. I mean, why would you ever not go with an individualized therapy if you're getting that kind of unique response to the INT. So how do you guys kind of see this developing obviously, near term, it will be whatever it's approved for, but longer term, when you have multiple approvals, where do you think INT might be appropriate versus shared antigen.

That's a great question. So a couple of comments I'd like to make about that question. One, we need to learn more about the populations that really would benefit from a personalized approach versus a shared approach. And we're learning more about that, and we'll learn more when we have our Phase III data and when we have more data on 4359. I think there could be some populations that might benefit from one versus the other. But of course, the obvious challenge with INT, as amazing as [ JAR ] is on getting that turnaround time to happen, there are patients with bulky metastatic disease who are diagnosed who need therapy immediately. They can't wait even 6 weeks to receive INT. And I think for those patients, it's really important that we have some off-the-shelf options that we can deliver to those patients and they can get the benefit of that therapy immediately. There's also a world where you can combine these therapies. You can say, hey, let's get the benefit of both to these patients. Let's give them access to an individualized treatment as well as start them off on an off-the-shelf product. And that way, they could have a better chance of efficacy from these programs. So I think there's a series of potential scenarios that could play out, and we'll have to make decisions based on the ongoing data that come in. Anything you'd like to add, Rose on that? Okay, or Jackie?

Alex Hammond, Wolfe Research. Just one quick follow-up on RSV. Do we now have a correlate of protection? And then on the Lyme disease vaccine, I'm very excited about that. Would you expect this to be a vaccine you're going to be getting it like every couple of years, every year, so how should we think about that revaccination?

Yes. Thanks so much for the question. So starting with RSV and the correlative protection. So I think at one of these previous meetings, we actually talked about the work that we did to establish the correlative protection, which we had previously presented at ACIP and it's actually been pretty well received by the regulatory authorities. It's how, for example, we were able to expand the indication to 18 to 59 year olds. So we actually are utilizing that work a fair bit. And as you saw, the immunogenicity data is really the booster data that we've generated in order to unlock that potential indication. Second question on Lyme disease and how we're thinking about dosing. So I'll say it's always dangerous to compare assay to assay. We know that assays can be different when in different hands. But we're seeing incredibly robust antibody titers relative to some of the competitive data that are out there with all of those caveats in mind. And so our thinking is actually we want to explore in Phase II a bit how we can reduce that dosing, particularly in the primary series as much as possible because in that first offseason, 3 is a lot of doses to get into place, and there's not a lot of wiggle room for people to miss doses. Subsequent to that, I think it's a similar story to RSV or noro as I've been describing, it's going to depend a lot on observing what happens in the real world. But we would imagine that the booster actually could be administered as a single dose, just like with other boosters, and that could be done in advance of the season, obviously, at a different preseason time than currently, but it's -- the way that we're thinking about it as a seasonal vaccine, that may actually be able to go a few years in between needing revaccination.

[ Woody Bogle ] from Bernstein on behalf of Courtney Breen. Just wondering if you could talk about the program discontinuations announced this morning. Aside from TMB, what was the rationale and what was the framework you used to make the decision? Was the data not sufficient, or are there still options for partnering?

Maybe I'll start and ask you guys to fill anything I miss. For the most part, it's -- product by product will have different answers, but the broad strokes are as we look to prioritize and drive towards breakeven and we look at the next step investment often in those programs, a large several hundred million dollar Phase III, we have a specific bar for the return on that investment. It has to really contribute meaningfully to our growth trajectory, and we have to view it as more valuable for the long term than other programs we have in our late-stage pipeline, across all of the pipeline, including oncology, rare diseases and what we're doing in the vaccine space right now. The ones that we discontinued in different ways for different reasons didn't meet that bar. They didn't feel like the commercial opportunity justified the level of investment necessary for us to do it. Even if we viewed it as value creating, we have a very high bar as we drive approach breakeven, as Jamey has reiterated again today.

And maybe I can just add, I've been actually asking for some of this clarity on some of these programs for a bit because as Jamey mentioned, if we have to cut in R&D, once we go into human clinical trials, there's actually a burden of work, regardless of whether you're actively vaccinating or not. And so somewhere where we're winding down, and it's not clear that, that's a program we would prioritize. Actually, it really helps me manage the workload and the staff if we can just stop. So I think that was another piece of it is to prioritize within latent, what we would go and do next, which I think you heard from, Stephane, is very much EBV in line.

Okay. We'll take our last question from Myles.

Just on the 1010 vaccine efficacy data, looked pretty compelling, quite provocative versus standard of care there. So there are some safety considerations, obviously. But just wondering how you see that as a commercial opportunity if you did have a claim to superior efficacy to what we're taking today, especially in such a large market.

Yes. We -- that's why we highlight that as a really '27 growth driver. Now we hope we're filing now. We hope to have that product approved, if everything goes well, in a year, but we'll probably miss the 26th season, but we really want to step into 2027. We think it's an enhanced profile. Ultimately, regulators will get a chance to review that and offer their perspectives. And we want to launch it globally. And so you'll see us in the United States, which is obviously an enhanced market, but actually across Europe, as I tried to highlight as well, we see this a big opportunity. And then as I alluded to, it is built into our contracts, our strategic partnerships with U.K., Canada and Australia and beyond. So it is a -- we tried to do 2 major growth drivers in each of the years. And the one place we allowed ourselves a third is in 2027 because it is between Europe, between expanding in Latin America through that partnership with Brazil and flu, we think there are multiple shots that get us there.

Thank you, Stephane. So this concludes the formal part of our Annual Analyst Day. So thank you so much for coming. I suggest we take a 10-minute break to have a coffee outside. And then for those of you who can stay, we have a team that has some interesting AI demos for you. Stay tuned. Speak soon. Thanks. [Break]

Can I suggest we get started for the AI section of the day? Team, can we get started? Thank you. Good. So thank you so much for all of you staying here and those online to look at the interesting work that our team is doing. Just maybe a couple of words of intro. Moderna AI journey started actually a long time ago in our science team. They actually convinced me back in 2016, '17 of the power of AI when it's used to tackle important product problems. They basically were working in the science to try to invent new enzymes using machine learning system that they had built with our data scientists and our biologists to improve the capability of enzymes so that we could reduce manufacturing cost, reduce purification costs. And the first time they showed me the enzyme that they had invented through machine learning system, most people were extremely not convinced, as you can imagine. But then they did the work. They made the enzymes physically that the system told them was going to be better than the enzyme that exists in nature. And the enzyme that was designed by the machine learning system performed as designed. So that was a big aha moment for me to know that an enzyme that did not exist in nature, that came out literally of a computer with a lot of smart biologists behind it could do exactly what it was designed to do. And so at that time, believing that AI was going to be an important feature of our work, we started Moderna AI Academy. This is pre-pandemic. And we started training people across the business to understand what machine learning could do for them, whether they are in HR or they're in finance, manufacturing, of course, in the science team. When you fast forward to, of course, the pandemic, we are kind of busy with the COVID vaccine. And then, of course, November 2022 with ChatGPT. Many of us played with ChatGPT over Christmas and had a little bit of a brain explosion and trying to understand why -- where this was going because it was clear that the November '22, December '22 ChatGPT was the 1.0 version. You don't have to have a lot of imagination. And we saw it ourselves we've seen the mRNA ride we saw over the last 10 years that if you have more compute coming, stronger systems and more data, the technology was just going to get better. And so we decided to really have Moderna starting to really incorporate into our culture and our work ChatGPT. We are, of course, very worried initially about, of course, not want to teach the rest of the world our stuff. So we developed mChat, which was basically a version of ChatGPT just for Moderna, totally siloed off from the rest of the world before GPT Enterprise existed. Actually, GPT Enterprise came from a lot of discussion we had, like I'm sure other companies with OpenAI. So win the solution for enterprise, what we can do on the Internet doesn't work for us. And then, of course, GPT Enterprise happened. So we stopped mChat and we really moved the company to GPT Enterprise, giving full access to people across the company. And I have to say when I go to see my colleagues, whether it's in R&D or in manufacturing or in HR or the GPT, we use and I use, I'm very humbled by what's happening. It is really, really exciting. It is just the beginning. But it's interesting to know, and I'm sure the team will share with you the number of GPTs we have across the company, and our people are just changing their work, which is why as you know, I asked a year ago to have Tracy leading HR and tech because I believe in the next few years, our job is to reinvent work across every line of work to go back to what are we trying to do for our customers and figure out what do we do with humans, what we do with digital systems, what we do with AI and what we do robotics. As you saw in the [indiscernible] facility, there's a lot of robotics work. There's more coming. And as the price of robotics come down and the cost of deploying robotics, thanks in part to AI for validation, I think this is going to be a very exciting journey. So with this, we're going to start with manufacturing. You're going to see us going through a lot of functions, little snippets, but you'll see actual live demo by the teams to build those things. And I think we're going to finish with science, which, of course, is where we think we have the most impact long term because everything else is mostly productivity, improving quality, which is really important, and it's part of our cost journey. And we'll finish by the icing on the cake, which is how can we do more amazing science, adding machine learning capabilities to great scientists. So with this, I'll turn it to Jeff. Jeff...

Good afternoon, everyone. Thank you very much for the opportunity. Thank you, St phane, Jerh, Tracy and others for the ability to not only speak about our work, but just to innovate every day using our large language models, our automation, our robotics. It's very fun to come to work, and I mean that truly. I'm Jeff Savard, I lead the INT manufacturing team based out of both Marlborough and Norwood, Massachusetts. Today, I have my colleague, Jason Manchester, who leads Norwood drug substance manufacturing based out of the Norwood, Massachusetts site. And today, we're going to be talking about use of LLM and AI within a CMC manufacturing environment. As a refresher, we intend to improve our profit margin by 10 percentage points over the next 3 years by 2028. There are 3 levers that we're pulling to be able to do this: volume, procurement and productivity and waste reduction. GPT manufacturing touches on all 3 of these pillars. In fact, we are a key enabler of executing this strategy. We can increase our speed and productivity to decrease our -- or maximize our utilization rate and decrease the amount of plant time we need to do things. We can create environments through automation and AI to have digital right first-time execution, to increase our quality and minimize our waste. And lastly, we can use digital and physical tools to physically miniaturize the batches that we're producing to be able to do more with less to ultimately manufacture in the smallest footprint possible. So we are a key enabler to the strategy, and we're really excited to be on this journey with the rest of the company. Within a plant structure, we have created dozens upon dozens, probably hundreds at this point, of custom GPTs, that span end-to-end production from raw material receipt through finished goods release. These are used and deployed by hundreds of employees daily in a clean room operating environment and supporting environments in support of execution of not only the core manufacturing process, but also are supporting quality and business processes. And these individuals are using these tools. They come from all spans and walks of life, from our operators to our entry-level associates, to our technicians, to our supporting technologists and our scientists as well. So we've truly created an environment where we have a digital-first mindset, and we are automating and enabling automation wherever possible. Today, we are going to talk about 2 of the pillars that are featured here, that being production controls and quality management and inspection. And we're going to walk through an actual problem that an entry-level associate might face in the middle of the night in a cleaner environment, where they have to act quick to be able to salvage the quality of a batch. So it's a real-world application, and you can see how use of the digital tools will enable faster resolution, lower downtime or less downtime and ultimately, increase our consistency and increase our quality. And what you'll see is the use of chain GPTs to be able to not only troubleshoot and diagnose an issue, but to assess the impact and then to run it through the full documentation cycle as well. So it's a full end-to-end and it's ready to be identified. So we're quite excited about that. So with that, I'll turn it over to Jason to provide the example.

And to just add on to what Jeff said, with the second and third move block here, the P-value GPT and the PIA GPT, these, to Jeff's point, with the dozens. I think we have 450 or so in total GPTs. We have built a suite of tools that we can essentially plug in place. So regardless of what the defect is that is observed on the manufacturing floor, whatever the event is, we can swap out those -- that second and third block down to cater to what we need to respond to the event. So I don't know what just happened here, but trying to get into AI with the GPT for the demonstration. Perfect. All right. All right. So what we have here is a live view of an issue that we had in our manufacturing process where we experienced a flow rate excursion. So this is something that it's not a common occurrence. It could happen at any point in time within our production schedule, so it can happen overnight on a weekend, on a holiday within a 24/7 manufacturing facility. So what we have here is an operator interface with the GPT that is specific to the area in which this individual is operating. They've stated the problem that was observed, that we had a flow rate excursion, that our manufacturing process went into a hold state, and our pressure is high and it asking what we do next to restart the operation. So we see quite a thorough set of responses in terms of coaching the operator to next steps, what physical steps to take place, what to look out for upon restart and also required escalation points. So who engage, whether it's a manufacturing science engineer or a quality associate, that is all covered within the immediate response. This is built with years and years worth of real-time experience, real life experience at the Moderna manufacturing facility, where our MS&T engineers, manufacturing science engineers as well as our process engineers have meticulously documented years worth of real-time experience with defects, problematic events in terms of equipment performance and resolution that was required for those events. All of that has been built into the background of this GPT as well as books of knowledge, we can call it, for the know-how that each of the individual engineers bring into Moderna. All of that has been codified into the brain for lack of a better description of the LLM here. So what we see here is a couple of interactions where the operator is continuing to state what we're seeing that we're going to be opening up a deviation. What we've looked at and confirmed is correct. So it's furthering the aim of the GPT to better inform the operator for what to do next. As you can see, there's quite the list of things that we can continue to look at. And this iterates a couple of different times where the GPT correctly points us to look for auditory cues, so unusual noises chattering on the pump that is controlling our flow rates. For this instance, that was truly the problem. So the operator asked for a bit more of a description around the chatter, suggests that they remember hearing something. So they want to look out for that in particular when we restart. So with this, the GPT informs the operator that we are in a safe from a quality perspective, position to restart, and it gives more of a know-how in terms of what could cause the chattering and what specifically to look out for. So from a mechanical perspective, roll aware for this pump is an issue, and that's exactly what this drills down to. So next steps here, we restarted, the sound was confirmed, and our GPT helped us to message the escalation to the maintenance department that would be coming to help us with that pump replacement as well as a batch record comment, which really kicks us off from the quality life cycle. It is taking all of the information that's been observed, then documented within the chat and giving us a GDP compliant right first time comment for us to put in the manufacturing execution record. From there, considering where we were in a process, this was part of a process qualification run. So one of the second and third block that I referred to earlier. We utilized this to chain in the process validation expert GPT to assess the impact to the qualification campaign, which then gives us an output that we're able to archive for this and continue to use as we get further into the quality life cycle. So it gives us quite a lot of information here. What this is built with is all of Moderna's documentation, all of Moderna's SOPs as well as years and years' worth of experience from the FDA, 483 guidance, things of that nature. So everything that we need to look out for in a GMP manufacturing environment. We continue on to chain in our product impact assessment GPT, which again utilizes all of the know-how, the experience from our MS&T engineers as well as all of our process documentation -- process development documentation to assess the issue that we observe to see if there is actually product impact. With that, it gives us a product impact statement. And we're then able to change in a deviation assistant, which then in almost instantaneously gives us a full deviation write-up, which can be seen down here. And the final chain in this, it was an audit preparation GPT as it uses canvas, so it slides the rewrite down, but we interacted with a GMP audit support assistant that we call Verify AI for us to ensure that it is FDA ready. So the deviation report meets all requirements, impact is solid, 483 history has been reviewed against it. This is a red flag or a tripping point. It gives us feedback. And now we're able to tie right back into the deviation guide to rewrite the assessment based on the report out from that audit support tool. So end-to-end, products identified on the floor, this event identified on the floor. This actually happened on the 12th of November, so it's relatively recent. It did happen overnight. And end-to-end, this is able to get to a full deviation write-up, technically speaking, within seconds.

Overall, within, of course, 5 to 10 minutes, an individual -- an entry-level associate was able to use essentially what would have been 5 to 7 phone calls over the course of hours in the middle of the night to be able to appropriately diagnose, troubleshoot and continue on with execution. So just to give you kind of a sense for how this will scale and how it scales every day. Thank you very much.

My name is Craig Kennedy. I head Global Supply Chain for Moderna. As St phane said, when he introduced -- GPTs are used as cognitive amplifiers all through Moderna today. I can attest that all through the organization, GPTs are deeply used by employees every single day. We actually wanted to show you some use cases outside of GPT today. We have 5 use cases in the supply chain area that we're going to show you. Firstly, we're going to talk about the way we use robotics in a very rapid solution for our last mile distribution. We've got some interesting things to show you there. We're also going to talk to you about how we use IoT devices and machine learning to track everything that we send in our critical markets to make sure that they get their on time and the right quality. We're then going to talk to you about a use case, which takes those data that we've gathered from everything that we've done, uses really forward-looking machine learning to create agentic capability to manage those shipments without the need for human in the loop. We're then going to talk about the way we do global logistics and global control tower using machine learning to make sure that as we move goods around the world, that they get there in the condition that they meant to at the right cost, at the right value for the customer. And finally, we're going to show you a use case that we use for machine learning for forecasting the distribution planning to make sure that we actually have exactly the right product in exactly the right location at exactly the right cost point all the time. Those are the 5 use cases that we're going to show you here today. Jamey is going to start off for us with robotics in the U.S. market.

I'm Jamie Collins, I'm responsible for last mile distribution in our U.S. market. So at the end of last season, we were presented with a challenge that we had to be fully FDA compliant, which means the PI, the product information leaflet, must be on or inside the carton, the salable unit. Given our current manufacturing environment and the time line required for that, we could not put it inside the carton. So we realized we had to have a solution at the last mile closer to the customer. So we did something a little unexpected, and we actually applied the PI on the carton. To do this, we enabled a -- and went on a robotics and automation journey alongside our partners, UPS health care and Eclipse Automation, and created a solution that would ultimately enable cost savings, mitigate labor risk and enable a scalable future for this type of activity. This project, normally, at many other companies, would have taken at least 18 months to implement and deploy. Here at Moderna -- and our team was able to do this in under 10 months. So from concept and design straight through to build and out to production, we built an 8,000 square foot system that sits at our UPS health care facility in Kentucky, that ultimately yielded us a 1.5 cartons per second throughput. So it's a significant automation journey there. And as I mentioned, we did something a little unexpected. And through our mindsets in being truly bold, it ultimately took a constraint that we had last year and turned it into a true competitive advantage. And what that means is that we are actually able to claim being first to market this year. You can see that's my colleague and I actually hand delivering the first doses to market for this season. We have a little video to show you of our actual robotic solution. Yes. And then if my colleagues could help kind of give you a little insight of what these PIAs look like. That's the next one. These are examples. Yes, excellent. So you can see there that our STAR robot decanting product out of the cases and putting them on to quite a conveyor journey there. And this is actually our next big product. This is a video of that. This is our new launch for this year. This is what we call our banding system. That's actually what was able to band the product that you guys are holding. All the while, something really critical about this system is that given our cold storage requirements for our product, we are in a frozen condition. We had to maintain time out of refrigeration and freezer. So this system tracked all the way across the 8,000 square feet, every step of the journey for each of these cartons to ensure that we are fully compliant within our handling requirements. And ultimately, this -- you'll see, there really weren't too many people involved here. And at the end there, again, the case gets closed, scanned, and you'll see one person there at the end taking it off the line. So it truly was quite an experience and something our team is truly proud of being able to meet product on time. In fact, an interesting stat is with this solution and other enablements we made this year, we were able to, after product release, deliver our first doses in under 8 hours to customers. So as you can see, something that we're really passionate about is continuously improving our customer experience, obviously, getting the doses to patients is of our highest priority. And so as of last year, we started implementing on a journey to enhance our customer experience by developing an industry-leading real-time shipment and monitoring solution that utilizes these tag and track devices. And what you'll see up here is actually what our customer receives. These are examples, again, not with our product in it. But these are examples of what our customer receives in the frozen storage condition. So the tag and track device is in there. And what that's able to do is provide us not only with location data so that we can ensure on-time delivery, but it actually provides us temperature data as well. And that's critically important given our cold storage conditions, et cetera. And what this has -- this year enabled us to do is create an automated intervention notification solution, which essentially, although it is manual today, but it allows us to be proactive by projecting where the product will be at any given time and potentially reduce temperature excursions ahead of it reaching the customer. So the cost-saving initiatives, but also a customer experience improvement. We have now shipped over 75,000 units -- shipments with this solution, and therefore, have gathered quite a bit of data that will then give us the ability to build off of that as Craig will then lead us into.

So one of the things, as Jamie said, we track every shipment. We've shipped over 75,000 already. We track every piece of data from every shipment that we send out. What that has done has given us a huge amount of data to infer what shipments are going to be good and what shipments may not be good. Essentially, it's a fraud detection problem. You have a very, very small number of shipments that don't go well. But when they don't go well, they are a huge impact on our customers. Particularly in a seasonal business, a vaccination provider, your CVS, your Walgreens or any clinic, when something turns up in one of those busy pharmacies that isn't ready to use, that's a problem for them. Yes, they get new products, yes, we fix it, but our goal is to never pass on that problem to the customer. The way we're doing that now is we're taking advantage of these data that we've collected. We've already built highly accurate inferential prediction models that allow us to know when a shipment is most likely not going to make it in the conditions that it needs to make. The reason that's important is because we can predict that now before the shipment arises. While we're in the process of demonstrating is an agent that takes that inferential data, listens to these devices which are here, they send every shipment every 10 minutes, that agent looks at what is happening compared to what we know is wrong. If it makes a determination that, that shipment is likely not to end in the right conditions at the pharmacy, it is built to do a couple of things. Number one, instruct the carrier to redirect the shipment. So bad stuff doesn't turn up at the patient -- at the customer. Number two, place a replenishment order such that the order goes out high priority immediately. And number three, tell the customer, all automatically, all on its own, so it doesn't have to wait for human in the loop to actually make that occur. We're even looking at the opportunity to determine whether or not we can segment and take higher value customers and make sure they get preference if a lower value customer is in route that we could reuse as well. But it's because of the fact that we took access to all of these data that we've collected, with the push that St phane and the EC and others have given us around machine learning and AI and actually being able to turn something into what will be agentic behavior as well. And that's good for the customer. It's also good for our actual costs as well because we can tell, even if it's going to excursion, whether it's going to excursion within the non parameters of what good product will be at the same time. So we save that return as well. That's something we're working with right now. We're very focused on the U.S. market. But what [ Gerhard ] is going to tell you about is also how we do this worldwide and give you a demonstration of our worldwide logistics control tower, too. [ Gerhard ]?

Yes. So similar to what we heard from Craig, we tracked our shipments since, let's say, I would say, the last 5 years, the challenge we had there is these were in different systems, fragmented data, nonvalidated data. With the control tower, we implement a tool where we through APIs, can ingest data from carriers from devices like this from devices like the tag and track here, but also from packaging providers. So the first time we are moving into structured data, validated data and an intelligence layer. So this is how we move from within, let's say, 3.5 months from selection in March, we wanted to be ready to supply into the U.S. market intercompany on our minimum vial product [indiscernible] and our Kuehne + Nagel shipments in July. So this is something we have achieved. And if we quickly go into the tool that you see a little bit how that looks, we see here when we go to -- if you go to the life shipments and then to my favorites, then you can see one of the first one. I can take the first one. This is now an intercompany shipment where you see, the shipment is going here from Belgium, where we have our central hub, which is feeding into the different country locations. It's rooted through the -- this shipment was going to Newark. And from there, it's trucked to Kentucky. So if you quickly look at the devices, we see here, we have here perfect data. We know exactly what the outside condition was, but inside is perfect in minus 20. So you see this is on track in green. What this tool gives us is actionable insights already. If something would happen and if there is a delay or there's a temperature excursion we are notified. This is the first step what we have here. So we implemented this one, but then we got the feedback from the market that what the customers get in U.S. for these nice boxes, what Craig and Jamey explained is predictive and an alert system they wanted to have it for the pellet shipments as well. So we quickly pivoted, use the same device for the truck shipments, and that is how we can go into the next treatment. That was the one we had opened before. So we use it also for the shipments out of UPS to the wholesalers, and they get an e-mail 2 hours, 1 hour and 30 minutes before expected arrival time plus a onetime link. They can see exactly the same here. And what we did as well, we have a geofencing solution. We have a light sensor in there. So whenever they unload it, it's automatically switching off the device, so we don't get any false alarm. So this is also what we then short note is implemented to really have control over the shipments, but also give the customer the insight and they can plan better the arrival of the shipment. And then the last one, what we did, where we saw a need is our critical release samples. We need to send to national agencies like CBER here in U.S., we need to send a release samples, and they are super critical for us. So that's why we implemented, if you go to the last one, [indiscernible]. We also put together with our partner, QuickSTAT, it's a white glove courier. We just use their API. We use their device. So we are agnostic. We use their device. And we -- yes, no, this is if you go back, it's the one with W. So there, we see that we have -- the same -- it's the one with W on it. And if you go to the device as well, please, you see exactly -- this is routed now and you see here the routing is not going directly from Spain to U.S., it's going through Frankfurt. That is on purpose to mitigate the risk because what you can see here if you are in Frankfurt at the airport, but also in U.S., we can store it at minus 20. So we mitigate the risk in case of an FDA or a custom hold that the product is getting out of range and we lose it, which would be incredible unfortunate for the guys then delivering to the customer. And what we built on top of that, we built the first robotic process automation. We have a central sample shipment or sample tracking tool, where previously someone manually had to enter the actual status of the shipments now with a robotic process automation. This is done every 2 hours or every 3 hours daily. We remove manual work. What is next year? So we are now 4 months. We have a lot of learning. The next step, what we want to do is automated temperature release. We want to take the validated data and enable automated temperature release feeding directly into our quality management system. That is one thing. And the second one, we want to do predictive thermal modeling. So using machine learning and AI to say, we are not static on a box and now it's validated for 80 hours for 90 hour, no. We know that if it shipped in 2 to 8 or in minus 20, it's much lower. It gives us more flexibility and we can improve our cost bases because we don't need to urgently react on each and every delay because we have the confidence and the tool tells us how long it lasts. What is midterm and that is then the real fancy stuff, I would say. And if we can go to the beta version here, we want to move from a static dashboard. It's really on a conversational partner. And you see here already the first preview, which is planned for mid of next year, where if you click on logistics, for example, so you come into the office in the morning, you can talk to the tool. And the tool took already the actions for you. No matter if you're there, it's not only then giving you the recommendation, it's taking the action for you. So it's escalating, it's accelerating or even like Craig mentioned here, it's returning. So this is then also where we feed in our lane SOPs, our risk assessment or even let the tool too for us the root cause analysis and the corrective action. So this is planned for mid of next year and this tool shows as well. We selected the tool in March. We started implementation in April. We went live in July. We added in end of July and beginning of August, these additional solutions for the market in U.S. for the CBER samples. We added the robotic process automation in September. And now we hope that in December, we can do the first automated temperature release.

Thanks. One final show, Joe is going to talk about the way we do distribution forecasting effectively by starting at forecasting likely shots in arms as a way for us to make sure we have the inventory in the right place, in the right condition at the right time so that we never miss a sale and that we have it at the lowest cost. Joe?

Thanks, Craig. So what you see up here is a live dashboard of a machine learning algorithm that we have implemented here at Moderna. This is dummy data. So there is not real data associated with this, so there's no inferences to be drawn here. But this is a probabilistic statistical forecast with machine learning backbone. And when I say machine learning, because we're not just using the projections that were developed from this statistical forecast model. We're actually ingesting real-time data, both from shots in arms administered, data we get through public databases as well as some internal data that we collect. As well as some -- and you can see here on the right, Google trends, right? There's some other exogenous factors that we track that are correlated to what our shots in arms and administrations are going to be. And so you can see here that we're looking at forecasting, we're not just looking at just a single point forecast, right? This is a probabilistic forecast that helps us continuously learn about where and how many shots we're going to need to get out to our customers. So if we want to skip data there. So this is just the U.S. market on aggregate. But if we go to the supply planning here, you can actually see we go down to the ZIP 3 level. So this isn't just about forecasting of what the total market is going to look like. This actually gets down into the local area of where we expect our consumption and where we expect our doses to be deployed. What that allows us to do is to forward deploy our inventory to satisfy customer need, and in many cases, before they need it. We also do get some customer feedback back around what inventory they're currently holding and so we can actually predict when they're going to need replenishments and make those recommendations out to our customers. So like we talked about before, we don't miss any vaccination opportunities for our customers, right? This helps to drive and ensure that we are optimizing where inventory is at the right time, at the right place and the right value to the customer. And again, this is a model that continuously learns. So as we get additional data on a regular frequency basis, we update these models. And as we think about how does this impact for our future, right? When you think about the connected supply chain, it all starts with what the customer wants and needs, feeding that back into how much we make, when we make it, where we make it and how we forward deploy that is a really big opportunity for us to optimize our supply chain.

Unless there are questions that's it for us. Good. Thank you. Right.

We save questions to the end and we'll move to development now.

Good afternoon. My name is Suzanne Tracy. I head the accountability for the transformation team within the pharmacovigilance area. We're actually going to walk you through how we've been using artificial intelligence within what is traditionally a highly conservative and regulated area. So about 1.5 years ago, because at Moderna, we do take advantage of artificial intelligence, we actually developed an AI center of excellence. Our vision being how can we revolutionize patient safety by putting actionable processes in place that are aligned with our operational excellence. How do we do that in a way that integrates our capabilities for artificial intelligence by building scalable, repetitive high-impact solutions. And how do we do all of this by utilizing artificial intelligence in alignment with our Moderna mindset, but to do so in a way that meets the regulations from the health authorities and remains aligned with our North Star patient safety? So we actually developed a foundational layer where we have originated an SOP a standard operating procedure for our software development life cycle. And on top of that, we've evaluated all the guidances from the health authorities. So the EMA, the FDA, along with other health authorities, have released guidance because they do see pharmacovigilance is a very data-intensive area and everybody is fully invested in looking at artificial intelligence. There are very few companies from our benchmarking that have taken it to the level that we have at Moderna. So we're going to show you some of those examples. With the talent that we have in house, Wen is actually going to take us through how we've used the compute platform and use that as a benchmarking -- can you go back to that slide -- to use that as the benchmark for our Workbench platform. So giving us an environment where we can develop our artificial intelligence solutions in a regulated and standard manner. And then on top of that, at Moderna, we remain invested in expert in the loop. So it's not just a human in the loop for us. In our area of expertise, we make sure that the human is actually a subject matter expert. And so all of this comes together so that we can enable all of our pharmacovigilance colleagues to focus on value-add activities. There's no sense in having our individuals at Moderna invest their time. And a lot of this does take an ordinate amount of time to review the data. So how can we do that more efficiently? So on the next slide, I'm actually going to introduce Wen. He's going to walk us through the Workbench, how are we using our compute platform to really introduce our regulated use cases. And then he will hand it over to Andrew. Andrew will actually walk us through very specific use cases that we've introduced in pharmacovigilance. So looking at our PV regulatory intelligence, using artificial intelligence to redact personal identifiable information. And then in alignment with the regulations, we have accountability to look at all of the social media with reference to Moderna. So how are we deploying artificial intelligence to use against that case? We're actually very focused on ensuring that our staff is focused on the higher-value activities. But along with these 3 use cases, we will be introducing savings in the millions of dollars. CSPVs, so our clinical safety and pharmacovigilance team is not lacking for ideas. We have 40 additional ideas in the pipeline. But because these are our highest impact ideas, this is where we've started this year. So I'm going to turn it over to Wen. He's going to take us through the Workbench.

All right. Thank you, Suzanne. My name is Wenhao Liu. I head up the data science and AI team at Moderna for the clinical development operations space. So let's rewind to early 2025. You guys have all heard of this term agentic AI kind of pop into mainstream as a buzzword. So we asked ourselves, is it real or is it hype, right? I think about a year later, it's mostly still hype in most places, right? And the reason is because it's not actually a problem of technology. It's an issue about company mindset and culture and it's the willingness to change and reinvent ourselves and really adapt our processes to build them up with agentic automation in mind. That's what it's going to really take to take us to this next level, right? . So I'm going to tell you about this platform that we built called Workbench. So this is one of our platforms that is enabling our AI-first strategy. So Workbench has 4 main pillars. The first is data, followed by data access in a secure and governed way, the agentic frameworks themselves and then the user interface. So I'm going to go through each of these one by one. Data. This is the most critical and most important layer. No matter what anyone says, AI will not clean or read your messy data. This is not -- it just doesn't do that. So this is where we found that most efforts stall because it's really a collaboration between people, process within the business and digital to take an existing business process that someone wants to automate. We trace the data to where they live. It could be in Excel, it could be in Sharepoint. It could be in digital systems like SAP, Syncade, Workday. We bring them into our data environment, and we map that data. We normalize it. We standardize it, and we push it back into this AI-ready data platform. So this is just a traditional data warehouse, right? But this step needs to be done and this is the really hard part. This part takes 3 to 6 months if the data is not in a clean state and you really can't do the rest of this without the foundation, right? I think this is where most efforts stall and because people try to skip this step and just try to get the agent to read the raw data. Then data access is extremely important in general, but especially so for a company like not just Moderna, but any pharmaceutical company, where the access to data has regulatory implications, right? So in Workbench within the platform, we built a custom role-based authentication system that is matched to our company's internal structure. And what that means is if I assign a role of medical monitor to a user within Moderna, that grants that person access to certain patient data, right? But it only grants them access to data within the clinical studies to which they are entitled to see. So that level of access control is foundationally required in a company like Moderna, in order to then attach our agent framework and have agents explore data within our network, right? So that is a foundational piece that we spent a lot of time thinking about and designing. And then you have the agentic frameworks themselves, which will access that data through the industry standard model context protocol. That's just how agents interact with data through a common interface layer. And then the last piece is the user interface, which is really critical because these agents don't just run by themselves in a dark room in the back, right? A user needs to be able to see what they're doing, right? They need to see what agents are in my team? What actions are they doing? What workflows are they running? Where are they stuck in a particular workflow and be able to interact with that agent in order to provide feedback, right, provide the expert in the loop and move the workflow along. So that observability is extremely important. And we are kind of limited in what we can do within ChatGPT. We've kind of pushed that to its limit, where we have done today, and we've done a phenomenal job doing that here. But the next level of this is providing a custom user interface, where users can interact with agents in a more structured way, okay? So I'm going to take you in through a demo of the Workbench platform. And so this is kind of executing what I mentioned earlier, which is the hard part is cleaning the data. So what I'm showing you now is a -- this is synthetic data, by the way, just to show you what this type of data the system has. We spent about 3 to 4 months working with the business within clinical operations and the finance team to take extremely messy Excel data from our CRO vendors to standardize them into a format, which allows us to look at clinical trial spend at specific clinical sites across our portfolio, right? It took us 4 months just to get this data in the system. And what I'm showing you now has nothing to do with AI, right? This is just a dashboard, but the foundation of it is the same foundation that will enable the AI use cases. So just to show you some of the information that this system has, what we're looking at here is data for a particular clinical study. So you can see the total study budget. The evolution of our contracts as they are executed over time. You can see the numbers are going up a little bit. So this combines our clinical operations data with enrollment and sites with our financial data, which is how much money we spent on these locations, right? And this type of dashboard will really give us insights. For instance, if you are looking at the number of active patients and number of active sites over time, and you're seeing this number going down, but you're seeing your contract costs going up. That would be a red flag as an example, right? And so what we really want to enable is instead of an analyst having to click through this -- these interfaces, applying filters going through every day and trying to find these insights. The agent has access to the same data, right? They can just do this for you autonomously. And that's the platform that we built, and I'll show you an example of that. So within Workbench, this again is a demonstration prototype. It is not live yet, in production, but it kind of gives you a view of the future. So the idea is that a business user can come into the system and create a financial assistant agent. This is done through configuration. They don't need to know how to code. The engineering team obviously will work with the business to build these agents. But once they're built, anyone can come in and configure this agent. So this particular one in this demo example, we are looking at clinical sites that have very low enrollment. So there could be a site that has little to no enrollment. And yet, we have upcoming site visits, which cost us a lot of money, right? So this will give the clinical trial operations leads insights into which sites they might want to close early. So you can see in this agent, we're able to configure the number of days ahead of time that we want to look. The -- what percentile of the bottom performance sites we want to look at, what therapeutic area we want to target and some prompt instructions, which can be configured, right? So this looks a lot like ChatGPT, but this goes beyond the capabilities of ChatGPT, right? This has much more customization and is much more fit for use for specific use cases. So once you can figure this agent, you can execute this workflow. So normally, once you configure it, the agent can run on a scheduled job. They can just run in the background, right? And when you come in, in the morning, you're going to get your insights and it's going to do all the analysis for you. But what I'm showing here is a manual execution of this workflow. And this little side window that comes up is actually from the collaboration with Open AI. They gave us early access to their front-end user interface component which we used in this application, but it's fully connected to our own custom back end, which has all the data level, security and everything that I mentioned in the first slide. And you can see the agent is able to autonomously go into our MCP server, locate the right APIs, make the calls, break it down into steps through chain of thought and do the analysis. So now this is the really important part because in ChatGPT that's where it would end. You would download the results from -- into an Excel sheet or you might read and do something. But because we're in a custom application, the output of that agent is actually put back into our database where the user can interact with these records right? So now the user can see, okay, I have a site here with 0 enrollment, about 13 upcoming site visits. The savings here could be potentially this amount, and you're able to open that record, interact with it, give the agent feedback, right? I can say this one looks good, I'm going to approve it or I can say reject, and then that provides the feedback into the system to allow it to continue to improve over time, right? So now if you just imagine, this is a generic workflow that we built, but imagine this apply to other areas, legal, HR, finance, manufacturing and you can kind of get a sense of the potential of this type of technology when you're able to put agentic AI against our internal company data. But I hope you come out of this with the appreciation of how hard this really is because it takes a lot of effort to get the data right, but most importantly, to align the people and the process and get the business team involved with digital. And I think Moderna, we have the right people, we have the right mindset. We have the technology for sure. That's the easy part, actually. But if you saw the presentations earlier, like just to give you the sense of the mindset of the people here, right? It's the people that are driving these innovations and it's the people that are going to make this change possible. And that's what makes this a really special place. Okay. So next, I'm going to turn it over to Andrew Semmes, who's going to talk about an actual use case we built and deployed to production using the platform that I showed you here, but only using a specific portion of this platform, which we were able to validate from a GSP perspective. So this is a really big deal. Now we can run agentic workloads in a validated environment that will pass FDA inspection, right? This is where we really get the benefits of this technology. So I'll let Andrew talk about the 3 use cases there.

Thank you, Wen. So I'm Andrew Semmes. I work in the AI Center of Excellence with Suzanne. I'll cover 3 use cases for you quickly today. So starting out with mScout. This is a regulatory intelligence agent. So biopharmaceutical companies are required to respond to various pieces of legislation and new regulations that breaks across the world. As you can imagine, monitoring 80-plus health authority websites, tracking all those changes assessing them against our internal procedures and then responding to them is a massive amount of work and is quite labor-intensive. So previously, this was a manual process, and we had a regular cadence controlled by business process where stakeholders would do this. They would have to open up their browser, look at the website. You can imagine that takes quite a lot of time across all the different regulatory agencies and the things that we're tracking. That's where mScout comes in. So mScout is our regulatory agent. So it runs on a daily basis. And basically what it does, it pulls down all the information from regulator websites it compares them from like the previous day to today. It looks for differences. It then translates from 25 different languages into English and then assesses that with a series of criteria looking for impact to our pharmacovigilance operations. So once that is running, you can see here kind of like under the hood, and this is part of the platform that Wen was talking about. It actually goes fully autonomous, right? So it has a trigger on a daily basis. And it will kick off its daily monitoring, look at the 83 websites and then send notifications to our experts in the loop who need to respond to that legislation. So here's just one example, right? The International Council for Harmonization recently had a new GCP guideline that come -- that came out. It's E6 R3. And in this case, you can see here for Argentina, the mScout went to the Argentina Health Authority website translated it from Spanish to English, identified the fact that Argentina adopted this international guideline and then sent this on to our LatAm colleague who's responsible to update our procedures according to ICH E6 R3. Then our expert in the loop. We'll take a look at the summary here. We have similar buttons, as Wen showed in the previous example, to indicate whether or not this is relevant intel or not relevant intel that feeds back into our model to track performance continuously over time and helps us understand when we make a change like upgrading the AI model if we're getting more accurate, less accurate as we go along. So that's just one example, but we -- and that one came out last year. This year, we're forging ahead, and this is one of the items that just recently came out. So when we process patient records, there are global privacy regulations that we have to comply with. So when a case has been closed, we then need to go into that source documentation, which could be on the order of 300 pages long and redact all of the personally identifiable information, names, addresses all of the above for PII. As you can imagine, highly manual, highly labor-intensive. We are also previously using a contractor to perform this work for us. By implementing this AI, we actually ceased that contractor contract, brought that capability in-house with our existing staff and achieved 80% time savings because we achieved in the 90 percentile in quality. So with that, I'll show you a demo of how this solution works. So you can see here, this is within our environment. We use a LLM that's internally hosted. So no personally identifiable information leaves our data environment. The user will upload the file. They'll get an e-mail. When it's done redacting, they'll then go and download that file. Once they download it, the AI is actually proposing those redactions, right? So you can see here and this is synthetic data. All of the patient name, race, ethnicity, address, phone number, e-mail. The user can then just look at the proposals. And if it all looks good, they hit apply all and then it gets redacted. So as you can imagine, this is a massive time savings across just one case could have hundreds of pages that this is needed for. Of course, we're a global company. So we, as you can see here, built it to be multilingual across 7 initial languages. This is a synthetic example for Spanish. So -- that's our second one, which recently went live. And I wanted to preview a look ahead of what's coming in 2026. So this is mDetect. And we are also obligated to monitor social media when someone mentions Moderna, mentions our products. We are obligated to look through that information and then see if there is any mention of side effects or product quality complaints associated with our products. We are also currently using a vendor to perform this. It's quite an extensive activity on the order of millions of dollars and so we're looking to apply that same strategy, right, have an AI agent that helps us go through all of that data, take a first pass at it. And then our expert in the loop with our existing staff will then confirm the disposition of the AI classification and either identify that there's no reportable information or if there is, send it to a downstream system for processing. And that will enable us in 2026 to also sunset that vendor relationship and bring that capability in-house as well. All right. And I believe that's our presentation.

We're now moving into the G&A section. So maybe one of the examples will be more relatable to everybody in the room.

All right. Hi, everybody. I'm Amanda Sorrento and I lead our HR core here at Moderna.

And my name is Nathan, I am in managing operations for HR. I just manage projects across our HR stack.

Excellent. So I'm really excited to be here today to talk to you about what we've been doing with AI and HR. And our journey kind of really started by looking at how we can redesign work within HR. And we started with some chatbots around benefits and around Ask HR that had great success in productivity. So we started to ask ourselves, how do we rethink the work within HR and in our core talent processes. And so about a year ago, we launched a GPT that is designed to support our year-end process. And so really helping employees as a support tool and then really helping to aid and enable our managers to have really great year-end performance conversations with their employees. And what we saw when we launched that was a ton of adoption, and we actually saw a net effect, while we can't directly correlate it to this, but we did see a threefold increase in usage of ChatGPT across the enterprise for different purposes after we had launched that tool to the enterprise. So we wanted to take it one step further from there and really think about now how do we aid in the development? How do we use ChatGPT and AI to really aid in the development of our workforce to build the capabilities, not only that they need today, but for the work that they need to do in the future. And just for a little context setting, traditional development planning happens where an employee sits down, thinks about what their development is, then they meet with their manager and they together come up with a plan. And oftentimes, that plan is only as strong as the manager capability or the employees' desire to think about their development. And so we really wanted to make sure we were building something that fit within the Moderna context, really helped to elevate what we're trying to achieve as an organization. And so that's where we brought in AI and ChatGPT, in particular, to really design an AI assisted process here. And so what we did was we took an API to Workday, where all of our employee data sits, and it gives great context about what's going on. It has the role of the employee as context. It has that employees' objectives, which are ultimately laddering up to what our organizational objectives are. And then behind it, we build an understanding of what the Moderna culture is, what our philosophy is around employee growth. And then how do we think not just about the skills and capabilities that an employee needs today, but what are those skills that an employee may need in the future, especially as technology evolves and all of our roles are rapidly evolving. And that became the structure for the GPT that we then launched to the organization to help an employee write a really quality development plan anchored in context and then also really bolsters the manager because it has a consistent context behind it and thinking to really ultimately produce a high-quality plan. This was a recent launch for us. So we just completed our cycle actually about a month ago. And so we think our outcomes are going to be measured over time and where we're really going to see the amplifying effect to this. And so over time, we'll really be looking at how does the capability of our workforce, especially in those skills in those areas that we think are critical for the success of the organization, how does that amplify over time? We think we'll see an increase in engagement, right? Employees want to know they're being developed. They want to have that commitment. And so we ultimately think we'll see a reduction, and we'll be measuring for a reduction in voluntary attrition. And then more exciting out of this is not just what it does around this particular process, but what it can also serve as we think about the whole employee life cycle and what more we might want to do with this. Because the data then goes back into Workday, which is our source of employee data, we can then leverage that data to think about what are the other employee solutions that we're putting in place to really bolster our workforce and drive for the company. And so things like internal mobility, we have incredible talent here. And so how do we understand where their skills are, what their motivations are and then really create a system from that where we can move our talent into new experiences and new needs of the organization, including roles that don't even exist today because they're still emerging as technology evolves. We've got a data-rich environment to be able to really draw from it in a unique way. We also can then really make sure our learning and development programs are truly anchored into the things that matter most. And we're focusing our energy with our employee base and what we're creating within HR to really drive those skills and continue to develop through the tools or through the programs that we may create as a workforce. So while those are our longer-term measures, we have seen some really strong good success just on the process, which tells us there's a high level of engagement. So we didn't design this with the intention of forcing everybody to go through. We designed it as something we wanted to draw the organization through into. And so what we ultimately saw was 84% of our workforce went in and completed their growth plan without it being a mandate or a requirement, which tells us that it really was a beneficial tool that we created. And pairing along with that, which really then points to the tool that we created was we saw 87% of our workforce utilize the GPT. And we wanted to understand, did they not just go in first time and then walk away from it because it wasn't value added, but we measured the number of conversations that actually took place, which told us they actually went through the whole process. And then ultimately, put it back into Workday, put the information back into Workday. So they really did find the value in the whole tool, and that's where we see 70,000 conversations that took place utilizing those tools that drove to that 84% adoption. And then obviously, one of the net benefits that we get out of this is it takes a whole lot less time for an employee and a manager to really come up with a quality plan at the end of the day. So I'm going to turn it over to Nathan now who's going to show you the actual tool.

Thank you, Amanda. It's going to get personal because we're going to do my own growth plan together. So it just -- yes, hang in there. So this is a pretty standard interface. I'm sure you know what ChatGPT looks like these days. And this is accessible throughout a variety of different ways. We had obviously some comms leading up to the launch of the growth plan. And we've got links available throughout our ecosystem MyModerna, which is our Intranet, et cetera. So you cannot miss this. And the usage rate just indicates that people have not missed this at all. And we've made this to be super, super easy. It's essentially a semi-structured conversation that we are leading people through and we're going to do that together. So I talked about -- the conversation is semi-structured just because you can say really what you want. We're not expecting much, really. There are some soft guardrails that I'll mention. Essentially, it's what you put in is what you get out in a more structured manner. So it's there to -- what Amanda was saying, recognize me, so my name is Nathan. And it will go and grab in the background but of information that knows about me including my role, my objectives as well, just again add a bit of flavor to what's to come. The growth plan is about kind of that macro lens for my personal development. And here, we have a pretty good overview of what we expect to put into Workday, which is the growth goal itself, some additional flavor on what is that growth goal and a start date and a completion date. This essentially helps us put on a silver platter what is needed for an employee and a manager to have a hopefully fruitful conversation about what's to come for 2026 and sometimes beyond given how macro sometimes these growth plans are. So here, again, it's asking me if I want to share if they want to pull my objectives, I'll say yes, sometimes the objectives need to be revisited. It's something that we prompt for in our quarterly connects and end of year cycles, for example. But here, I forgot. Oh my goodness, what do I need to do? That's right objective one, objective two, I'm all good. Very big objectives. But -- so I refreshed my kind of my objectives, and I can go and now starts a series of 6 questions. My Moderna colleagues, I hope are within the 87% of people who went through these, so they'll know them very well. And these questions have been put together so that they can offer our employees a very a best-in-class type of questioning. So it's a lot of self-introspection, again, leading to that manager conversation a few weeks later. So we start with like any good plan about the future, we start about a few snapshots about our current situation. So here I will -- I'm a rubbish typer, so I will go and copy paste what I have in another window. Very, very fancy. And again, like I said, it's not adding anything. It is going to regurgitate what I've put in. Like Amanda said, it's got the Moderna mindsets, our corporate objectives, a lot about our -- who we are as a company in the background. So it helps to format, a lot of my inputs into something that will be again the basis of that growth plan conversation. So again, summarizing what I've put in. I don't feel like I need to go further, but for each of the questions, it's going to probe additionally to really go and nag the usage to ensure that they're giving as much information as possible to the model. So I'll go and say no, and then we'll get on to the second question. Again, second question is mostly about current situation. So more of what has been going on about my projects, what I've had maybe some troubles with. I feel pretty good at the moment. Thank you very much. I don't need to add too much. Again, current situation, snapshot, core strength. So here, we really depend -- it's a big signal for our managers to understand if our employees know themselves. It's a big point of feedback as well. So that's a big important question that we've included here. And I feel pretty good about that answer as well. If you remember, I've mentioned it's 6 questions. So we are now halfway there. And that was a tough one for -- all right. We're halfway there. Now growth edge. This is where we start to slowly bring the conversation into the future. Where do I feel I can grow really. Again, I don't expect the GPT to add anything to the mix. It has that context of my objectives. It has that context of my role. I'm just kind of guiding it towards what I feel I could strengthen in terms of skills and kind of picks up on it. I don't feel I need to go further. And now really this is the -- we start to do a bit of soul searching here, what's coming up. So future horizon, what do I have in mind? Again, I have some -- I know exactly where I want to go, of course. So I have exactly what I want. And I'll just insert something here where you would think, well, easy, Nathan, I can game this and just say to my manager and [indiscernible] here, I want to get promoted and boom, 2026 promotion. Not quite as easy as that. So I also want to get promoted, rubbish text -- rubbish typing. And so here, it's taking what I've given it in terms of my 12 months objective, but also kind of nudging me to talk to my manager. When it comes to promotion, this is something that is, of course, part of working in a place like Moderna, but it is something that is not necessarily linked directly to my growth plan. So here, we're kind of nudging to talk to managers, to engage them one-to-one to ensure that we have something that is tangible in these growth plans. And so I'm good to go here. And so here, we come to the final question, which is the -- I think the most meaty of them all because we're really trying to make people think about the different ways that growth happens at Moderna. We have a pretty solid framework around how that happens. And so it's here exploring the ways that I think I can grow, adapt my skills for today, but also for tomorrow and the HR tech stack in my case that is coming up. And so I've, of course, given it a lot of thought, and I can input that in there. And at the end of the day, very, very easy output, right? Nothing mind boggling. It's going to give me exactly the different elements that I need to copy-paste straight away into Workday. Workday becomes the start of the conversation for you and your manager. Your manager can consult that into Workday pretty seamlessly, and you can have that one-to-one your manager in the moment that's in and hopefully smash it out of the park in 2026. And that's all she wrote. Thank you very much.

Thank you. Could you get it going and I'll start talking. While they're pulling up the slides, hello, everyone. My name is John Ward. I'm -- hi. I'm John Ward. I have to say I'm just humbled to be up here speaking with you today, and I'm really grateful for just all the great work I've seen that have been in the presentations that have preceded me. So I'm going to be quick today and to tell you about a GPT I created that I use every day to great results, I think. So I'm the Trademark Attorney here at Moderna. I'm sort of a team of one, as you can tell. And I think if we can bring someone to advance my slides -- so I deal -- Trademark Attorney in the pharmaceutical industry, which I've been in-house as a Trademark Attorney in pharma for about 18 years. And all of that time, I've dealt primarily with creating drug names, right? So I handle the legal aspect of drug name creation. It's one of my primary deliverables. And there's three components to a drug name. One is commercial, one is legal and one is regulatory. So you can go back up to the original slide. So the commercial part is -- it's a conversation that's a good name, right? So that's a very broad conversation. Everyone can weigh in on that. The legal part of it is the name that has to be, we can own the name, right? We can use it freely without stepping on anyone's toes. And the third one is regulatory, that is we can get permission to use the name from the health authorities. So as you go down to those three, it's a whittling down of names from a big bunch of names to a smaller bunch of names to an even smaller bunch of names. And I'm going to talk about that second step briefly how I use ChatGPT and an AI, a GPT I created to address the legal screening of names, right? It's search, it's called trademark search. So trademark search is the most expensive thing that I do. It's the biggest part of my budget, traditionally, the biggest part of my budget throughout my career. I've spent millions and millions and millions of dollars on trademark search over the years. But when I came here, I said -- I was actually hired already. I was talking to the General Counsel. And I said, "My goal -- everywhere I've worked, everywhere I go, I see waste. There's a lot of waste in this process." Unintentionally, I mean, it's just sort of baked in because you're dealing with -- you're going from ambiguity to quantifiable risks. That's what you do in this -- in my space, right? And in that process, you're going forward with a lot of uncertainty. There's a lot of dead ends but you only realize that after you spend a lot of money. That's regrettable and something I've always tried to address. So I said to the Chief General Counsel, Shannon, I said, "My goal is to come to Moderna and build like a virtual program that is heavily reliant on technology." And so that's fortunately, the technology caught up and I was actually able to do it because I don't know if I did -- would be able to when I actually said it. So here's the evolution of trademark search, right? So trademark search is, again, you go from this big, you're narrowing down a pool of names, hundreds of names to get down to about 20. And you try to figure out which 20 should you advance to stick into the regulatory review. And the traditional -- tradition in the industry, what I've done throughout my career is you rely on outside counsel because this is really -- it's really about scaling yourself, right? So you have to scale yourself out. It's too much work. You can't do it as an individual. So you need to send it to outside counsel. Here, we have very good outside counsel that -- I've built a team. It's a lot of commoditized work. I rely on counsel in Eastern Europe, very, very aggressively priced, great savings, but there's only so low you could go. And the low you can go is $350,000 a project, and that's per name for search, right? And we have five or more names in the pipeline, so -- projects in the pipeline. So it's expensive. In that you can get 20 names in about 50 countries searched, take you about 3 months. That's a very competitive price. I can tell you what big pharma pays for that, but I'm not going to. It's a multiple of that. So -- but -- in 2024, we're hit with cost constraints or constrained environment and resources. So I moved this work in-house. And I -- but I shrank down the number of countries from 20 to just our key countries. And -- but I could do it myself, fewer countries, 42 days, and I got a subscription service that allowed me access to global databases of trademarks. So I would basically do the job that here, 50 lawyers are doing this work, one in each country, that's how you do it. I'm doing it now, but I can only do it in fewer countries because that's just impossible when you're doing 20 names, 50 countries, do the math. That's a lot of searches. So -- but I drove hundreds of thousands in savings in 2024. But this is unsatisfactory, right? Because it's a compromise in the global footprint of the search. And it's not -- I want to have the industry standard, right? 20 countries -- 20 names, 50 countries. This is moving forward with too much risk, which is sort of what a lot of companies will say, "Well, we're cutting resources. We're going to have to like live with this." I don't want to live with this. So I created a GPT that could scale out my ability to deliver a bigger global footprint, more search, more search, right? So I went to -- I wrote a tool called [ M Clear ], in-house, relying on commercial search database. So I got my annual subscription fee, but I can do that per project for basically nothing. And this is driven -- I did the math this morning, so far this year, 7-figure savings and 7-figure savings in dollars. So that's all kind of good. Let's see what see if I can do this. I'm going to see if I can do a demonstration of this quickly, hopefully this time. So this is sort of like unlike the projects you've seen before, which are very impressive, this is sort of like Gorilla GPT, right? I just did this at my desk over the course of about 6 weeks, I wrote this thing. And I told my boss, "Hey, look at this, I think we can drive a lot of savings with this." And so then I implemented it. But like no one told me to do this or no one said it's okay or -- but I wrote it over 6 weeks, and it's basically Python code, right? So I just go into -- I use chat o4-mini. So we'll just start this, so what I do when I start this, I've got these -- I just built this folder. And this is all again, I guess, what someone called synthetic data or something. The name I'm going to search is not real. But -- so what I do is I -- this is what I call the [ M Clear ] engine, which is the real GPT and some 133 lines of Python code, each one a polished gem. So then I just take a search. This is an example of 500 lines. What's in this, what I just deployed is a spreadsheet. The name that I'm creating, say, hypothetically is Syngenry, that's a name candidate. So I want to know what the risk landscape is for Syngenry. So I search it in the U.S., I get 500 senior trademarks that might be a problem for Syngenry. But I have to figure out what's the risk landscape within those 500 names, right? That's hard to do, right, without -- because you have to go through 500 lines on the spreadsheet and then rank them and then understand them. So that's hard. So -- and that's what this tool solves for. But that's what you -- that's why you outsource it to lawyers. So here's my prompt. I'm just going to copy that here. Then I just say Syngenry. Guess I could have typed this beforehand, didn't I. And in the region, this is the U.S.A. And today [indiscernible] in front of a live audience. And then you hope you're connected to the Internet. And then you hit that and you wait a few seconds, right? So it's uploaded the spreadsheet of all those third-party rights. It's applying the Python code in the GPT. It's working in the background. For 5 seconds, it's analyzing. Okay. So here, it will upload in the window the data, right? Here's a preview, it'll give me a preview. You can download the -- all right, let's download it, click the report, comes up here, [ M Clear ] Syngenry U.S.A. Click on that. Right, so here is the [ M Clear ] report, and I'll go back to the slides because it's easier to see there. So there's the top row of the report that I uploaded. This is what's called -- I just think of this as the raw data. And this is just right out of the USPTO website, right, right out of the USPTO. And so to figure out what the -- to understand the risk -- it's a risk landscape for Syngenry in the U.S. To figure out what the problem marks are, and there's some in there, you have to read 500 lines of data and then keep track of what's the problem and what's not. It's hard, right? Do that 20 times and you've cleared your marks in the U.S. And some of these reports are 2,500 lines long. So here's the [ M Clear ] report that I just opened there a second ago, but this is whatever, data, text wrapped and the windows made the right size. So it adds -- the report adds seven columns that tell me what the risks are in the spreadsheet. First, I have whether it's a high-risk company, yes or no. And in the GPT, I've identified companies that I add to that list as I go forward, like CSL, major vaccine manufacturer and so forth. So this is very good to have. And because you can just toggle just to look at the high-risk companies because they're the ones that are most likely to create problems for you, your competitors. So I have an overall similarity score that just translates medium, high, it goes from low to high. And so -- and then I usually put filters there. And so you can filter for any one of these things. Let me see what the highest and the medium-high and high are. Then I've got weighted high-risk score. And then these are the algorithms that are running. This checks for spelling, like how many letters would you have to change in order to make the words identical. These check for two and three letter sequences. I've got a combined overall similarity score. Wow, okay. So I've got seven different things here. And these are all instantly done. As you saw, 500 lines, each one of these things was calculated in a couple of seconds, right? So this saves an enormous amount of time, saves an enormous amount of money. I don't pay for search in any country in which is a Latin alphabet, right? It's over -- it's over $1 million this year so far in savings and the money you save -- money you don't spend on outside counsel is the sweetest money you can possibly have, right? So and then I can go to R&D and other things. So that's it. It's very small. It's my job, but it's scaling myself out, which is exactly what I want to do. And it's part of a larger end-to-end AI solution that I'm working on because -- but with this is, is really the money driver. I broke it out and bolted it on to the traditional workflow that I've got. Great. That's it. Thank you very much.

And last but not least, we have research, and we'll start with protein engineering, followed by antigen engineering.

Okay. Thanks for staying and for all the interesting AI talks we have here at Moderna. This has been actually really informative for me too, what's going on. So I'm going to tell you today -- well, my name is Dan Kulp. I'm a senior fellow. I run a group doing computational design and in-vitro selection at Moderna. That sits in the Protein Design department under Bill Schief. I'm going to tell you guys today about how we're using AI in protein design. I'm going to sort of teach you a little bit about how we're doing protein design in the world of AI. And hopefully, it will be interesting. Just as a way of background, both Bill and I have been doing protein design for decades. And when I first started off in the field, and I told people I wanted to design proteins on a computer, they thought I was joking. And the field has grown tremendously since that time. However, both Bill and I have been able to design proteins on a computer that we've evolved into Phase I clinical studies. So we've been testing some of these things we've designed all the way through preclinical development into Phase I. So as we all know, Moderna makes mRNA medicines. The RNA goes into cells. The cells are instructed by the RNA to make proteins. So it's our job to make sure that those proteins have the right shape and right function so the medicines actually work. So we spend a lot of time crafting these proteins and trying to satisfy all the requirements of mRNA medicines for our proteins. Okay, so another way to put this is, can we choose a sequence of amino acids that help a protein fold into the shape that we want and perform the function that we need? And it doesn't sound like a hard problem. But if you start thinking about large numbers like the number of atoms in the universe, the number of chess games possible, protein design, even on a modest protein of 100 amino acids is a larger problem than that. So we've come up with sophisticated computational algorithms to try to solve these problems in the past, and AI has helped us go even faster. So basically, designing novel biomolecules was really slow and unreliable previously. But now we can sort of rapidly create these de novo proteins with specified structures and functions, and we're just moving a lot faster than we were before basically. Another challenge had been that we have low experimental success. It used to be that 1 in 100, 1 in 1,000, 1 in 10,000 designs would actually work for us. But now, which is -- the other day in my group, we had 1 in 10 that we order work because we're using these new AI methods. So we're basically able to go faster and build better molecules, which is pretty cool and allows us to be very creative because we can iterate faster and -- over concepts and design proteins that perform better. Another challenge had always been sort of this limited exploration of molecular structure. So these AI tools allow us to build complex structures. And so I always think of it as it's opening this universe and allowing us to dream bigger and be bolder about what kind of proteins we actually make. And so I think it's really accelerating our preclinical discovery and broadening our potential therapeutic reach. So I'm going to show you a few movies just to show you what the protein design cycle looks like. This is a common paradigm in the field that everybody is starting to use now for AI-designed proteins, and it's this four-step sequence. And each of these steps has a number of AI tools embedded, baked into it. And I'm just showing you the basic concepts here. The first thing we want to do is we want to design the fold. So we got to have the protein make the right shape. So there's new diffusion methods that take atoms that can be randomly distributed in three-dimensional space. And over the course of the simulation, can start folding into things that look more and more like the target protein structure that we're trying to make. And you can see the simulation, if it finishes, will come into a structure that looks like that. And so at that point, you have the fold that you're looking for, the shape you're looking for, but you need to design a sequence to go on top of that. So the next step in the process is a sequence design step where basically these new algorithms can just decorate these protein structures with amino acids that help it fold into that shape. And once these simulations are done, you basically have an amino acid sequence of a protein. You could go into the lab at this point and just go try to make it if you want to. However, there's been a lot of innovation in the structure prediction field that actually the Nobel Prize in chemistry is won for this purpose because tools like AlphaFold are so good at predicting a protein structure based on its sequence, we can use it in design. We can ask ourselves, in this design sequence that we just made in this pipeline, how well does that actually fold into our target state? Sometimes it says yes and sometimes it says no. And I'm just showing you an example here where obviously, it looks like the desired protein. The last step is a structure filtering step. This is because this pipeline generates lots and lots of designs, tens of thousands to 1 million designs, right? We're not going to go make all of those in the lab. And so I kind of think of this structure filtering as sort of the special sauce of protein design. Different companies do it differently, different groups do it differently. And so basically, you take a number of different designs and select out the ones you want to go experimentally test. Okay. Now you're protein design experts. I'm going to show you three vignettes of how we're using this design to actually design proteins at Moderna. So first is designing vaccine candidates, okay? So here's an example where we had these two proteins, this green protein and this red protein. And we thought, hey, these might be pretty good based on the biology of proteins that we want to make a vaccine for. However, when we went and made these proteins in the lab, they just fall apart. So they're not good vaccine engines. We had no way to make the vaccine. So what we did is we ran that cycle that I explained to you, [indiscernible]. And in the green case here, we built a de novo dimeric scaffold that actually stabilized that protein. And in this case here, what we did is we actually connected two parts of the protein with a brand-new piece of a protein. And in both cases, when we go make these in the lab now, these proteins express and are good proteins, and they behave really well. And further than just making the proteins, we've even done mRNA delivery of some of these AI designs. I'm just showing you data here where basically the gold standard in this vaccine is in these squares and our AI design is in these triangles and shift -- curve shifted over to the right are better vaccines. So basically, our AI design in this case was more potent than the gold standard. So we're pretty excited about that. Another class of proteins that we like to try to engineer are these proteins called self-assembling nanoparticles. These are proteins that are multivalent. And what's great about that is you can put your vaccine antigen onto these proteins and get multivalent display. Immune systems like proteins like this because it looks like a virus to that. And so they amplify their response when they start seeing multivalent proteins. So there's -- in the past, people have used lots of naturally-occurring self-assembling nanoparticles. But now with AI, we can actually just build these from scratch. So that's what we've been doing. We were just building these de novo nanoparticles from scratch. I'm just showing an example of one here. I'm showing you data confirming the assembly of this when we go make these proteins in the lab. Okay. So the last vignette here is this idea that there's a lot of interesting biological molecular surfaces out there in biology, and we may want to develop binders to them. One classic example here is a peptide-MHC complex. These are targets of T cells. And so it might be good for us to be able to design binders de novoed against these types of targets. So that's what we've been doing. Here's one case where we designed what's called a mini binder. We allowed the protein to fold into any shape it wanted to as long as it could bind the target. Where is my mouse. And we diffused it just like using that pipeline, and we came up with proteins that look like this. This is a three-helix bundle protein, but we can also ask for antibody-like molecules. And so we've also been doing that with just de novo design of antibodies. And we can see here that this is a case of one of our mini binders that basically this is a binding SPR experiment. And you can see that the on-target binding is very strong, whereas the off-target is very low. So that means a peptide-MHC complex that is closely related, but not the right one, we have no binding to. And so this specificity is actually really important for this type of application. But we're able to accomplish it through de novo design. Okay. Besides this, I think, really exciting protein design cycle that we can do, we can design proteins of all different shapes and functionalities. Our group is using AI in many different aspects of protein design. So one is using these large language models as scientific intelligence systems that know all about the research that's going on around the world and about our research and can help us brainstorm about what are the possible next steps. I mean we've made some of these proteins, what should we do next, given everything that's going on out there. And I think that's been a really great use of AI in our group. We've also built focused GPTs. So both computationally and experimentally, we're generating lots of data, large amounts of data. And the question is, what does the data mean and what should we do next? And so these GPTs are helping us as data analysts to analyze all that data and to help us come to some conclusions. Lastly, there's agent AI, which you've heard a little bit about these agents previously. And we use it for designing the code to build these protein design pipelines. I've been coding for many, many years, and it's always been a slow process. But now with these agents, you can actually just launch them and have them generate the code, test the code and deploy the code. And it's just been amazing that it can do all these things so well. I mean the software engineering has been a really great implementation of these large language models. And what's important is that, that cycle that I was telling you about on the first slide really needs to connect to a large compute infrastructure because we need a lot of computers to do what we're doing. And so these types of agents can actually build in that infrastructure for us. So why does everything I told you matter today? I mean, first of all, this automated coding allows us to really rapidly assess new AI tools. There's new AI tools coming out all the time that are amazing, have amazing claims. Sometimes they're great and sometimes they're not so great. Sometimes they will be good for Moderna's applications and sometimes they won't. So we need really quickly to implement them, and these agents allow us to do that. They build it into the pipeline really rapidly. I think also what matters here is that we're able to rapidly develop really high-quality drug candidates. So the proteins we're making now are just way better than the previous generation of proteins we were able to make. They're more stable, they're more specific. And I think that's going to just be better for developing medicines around the around the company really. And the last thing is, which is I'm sort of most excited about is this idea that you can dream bigger. It allows us to think about new concepts we could engineer going forward that we really didn't have the tools in hand to do at all. And so that's one of my most exciting things about AI really for protein design. So it's -- I have this tagline here. It's like a co-scientist. It's allowing us to sort of go faster, make bolder advances and really help biomolecular design across Moderna and hopefully bring medicines to patients faster. That's my last slide. Thank you.

All right. Thank you all for sticking around. And yes, we are last but not least, hopefully, I'll tell you some exciting updates. So I'm Kristine McKinney. I lead the cancer vaccines research team. And these are concepts we call cancer antigen therapies externally. These are pioneering medicines, and we have many in the clinic at this point, and they are all centered on and all designed by algorithms, which [ Wei ] is going to tell you a little bit about later. But both the personalized product, the individualized product, which is in partnership with Merck as well as the shared products, the off-the-shelf products are all designed by these algorithms. And fundamentally, the way these medicines work, right, is by hacking into a common piece of biology, which is that all cells in the body are displaying on their surface a sort of fingerprint. So there are pieces of proteins that are expressed in those cells and then displayed on the surface. So what our -- what these product concepts are about, the antigen therapies, is actually trying to teach the immune system how to recognize the fingerprints that are specific to tumor cells and not normal cells. So as you can imagine, making sure that you get that fingerprint right is at the heart of both the safety and the efficacy of the drug. And although the techniques for looking for these fingerprints are getting more sophisticated, it's a huge amount of data. It's complex data, it's noisy data, and it has a lot of features that bear both on safety and efficacy. So it's an amazing place for AI. And we've deployed it in order to make sure we're leveraging all of those data. And not only to actually discover those fingerprints so that we can design the drug, but also in terms of which peptides are present on tumors and not on normal cells, but also looking forward to the next step, which is which of those fingerprints can actually be -- are capable of being recognized by the immune system. So both of those, [ Wei ] is going to talk about in more details with no further ado.

I can stay here. Thank you, Kristine, for the introduction. So I'm [ Wei ]. As Kristine mentioned, I will walk you through this biological mechanism behind all of our cancer antigen therapies, including intismeran as well as off-the-shelf cancer antigen therapies. So as Kristine mentioned, the biological pathway for this antigen display and the T cell immunogenicity is very complex. First, your source protein will go through this transcription and translation process according to central dogma, but that's not the end of the story. After the protein is expressed, it's actually going through a really complicated pathway, depicted by this figure below, to display on the cell surface, which is that peptide-MHC complex Dan just mentioned. Once this peptide is displayed on cell surface on the MHC complex, it still needs to be recognized by T cells and trigger T cell immunogenicity, which is the holy grail that activates everything that is underlying of our drug efficacy. So today, I will talk to you about three things. First, what is the key technology that enables us to recognize these patterns on antigen display and how we're using AI tools not only to make this pattern recognition much better, data collection much accurate, but also use it to integrate the large volume of data to make predictions. Even if we don't do a very labor-intensive experiment, we can also predict antigen display very accurately. And after that, I will talk about impacts of this technology to our current pipeline programs. And at the end, I will talk about some forward-looking things about new data modalities that's emerging and new models we're trying to build. So first, I want to introduce this powerful technology called immunopeptidomics. Sometimes it's shorthanded as IP-MS, which stands for immunoprecipitation-mass spectrometry. Those are the two key steps used to acquire such antigen display data. After we acquire data using this instrumentation, we need to analyze this raw peptide spectral information. And right now, we are actually at a very exciting point, almost like Sanger sequencing to next-generation sequencing inflection point over 2 decades ago. For this IP-MS field, we're seeing the explosion of data volume. We're seeing the instrumentation getting better and better in terms of data throughput and quality. And at the same time, we're working with a lot of partners using the best AI tools in the field to be able to analyze this data really faster and with better quality. We actually did some quick calculations on how much better we're doing right now, just comparing to 1 year ago before we acquired this powerful instrument internally and deployed these AI pipelines. These days, we can actually acquire 5x more high-quality peptides that are coming through this IP-MS instrumentation. And we also got 65% higher-quality identifications just based on some key metrics in mass spectrometry field. And we are also doing two different data acquisition mode. One is called DIA, data-independent acquisition and layer on top with a more sensitive data-dependent acquisition. By combining this and using AI to [ rescore ] the target identifications, we get 1.7x more high-quality targets. Finally, we have integrated all these great tools that have small logos in the bottom pink box into an in-house pipeline that's fully automated and the processing time is much, much faster. When you feed it with a raw data coming from an instrumentation run, you can get the instant results after 2 hours, which is already state-of-the-art performance. And with that, I will show you two case studies of using this IP-MS technology and AI pipeline. On the left, I'm showing you a preclinical cancer antigen therapy program we call [ Lion ] because cancer antigen therapy is shorthanded as CAT and [ Lion ] is a very powerful big cat. We have very high hopes for this cancer antigen therapy. And originally, we designed a bunch of antigens to come into this drug product. However, one of the design, as you can see from the second flow cytometry part, it doesn't have the optimal protein expression features as measured by flow. We quickly come up with multiple redesigns and highlighted two very promising designs done, the design 10 and design 12, which showed promising signals on flow cytometry. But as I mentioned previously, protein expression is not the end step. We still need this protein to be digested, peptides presented on the cell surface. That's where the IP-MS really comes into play, where you can see with this orthogonal technology, we quickly proved that these peptides originated from design 10 and design 12 can be detected on the cell surface on the right, MHC complex. So those are the data we collected very fast with this internal latest mass spectrometer. And 3 days after we acquired all the raw data from multiple designs, multiple technical replicates, we were able to comb through the data within 3 days and inform a locked drug candidate to go through the next stage of preclinical development. So that's the power of wet lab and AI pipeline combined, new technology. And we also know that we're not the only one leveraging on this technology. In fact, the whole field has been painstakingly collecting such data over the decades, although the instrumentation was not as good as current -- right now. And there are a long-standing effort to try to predict what antigens can be displayed on certain HLA alleles on the cell surface. And these kind of models are very important for antigen design. And that's also -- this kind of public data is at the core as one component in our intismeran algorithm. So what we are trying to do here is to keep developing even better algorithms that fits our purpose. And in the real application scenario such as intismeran, we care about the very specific feature of such models. We need to make sure that what we call as most likely to be displayed antigens, they are truly positive. And using this specific positive predictive value metric, we benchmarked our internally developed AI model against all the state-of-the-art models, including some really well-known ones out in the field. And you can see that the pink curve on top is the model we currently have developed in the research environment trained only on the public IP-MS data. And I just told you that we now acquired this latest instrumentation. We have this very fast AI pipeline. So we are now collecting proprietary data. And we also have ongoing collaboration with Immatics, which is another expert in this field. So together, we're really trying to push this frontier for cancer antigen therapies and aiming to develop the best performing model. Now the last piece is looking forward, what's next in the field. Once we have a highly accurate antigen display model developed, how do we make sure that these antigens are really triggering T cell immunogenicity. Right now, there are many great thinking in the field, leveraging AI, leveraging different biological principles, and we have seen many, many TCR-agnostic immunogenicity AI models out there in the field, and we're also developing our own. Right now, we have some ready-to-use research models that we're investigating, and they usually rely on some antigen-level specific experimental data collected through your traditional immunogenicity assays such as EliSpot or intracellular staining. However, another very exciting new emerging trend in the field is that we're seeing more and more TCR-specific data emerging, where you will have a specific peptide on a specific HLA allele binded to a specific TCR sequence. This type of binding data were very hard to come by in the old days. But since the pandemic, we have seen Adaptive depositing millions of such binding data into the public domain, but that's in infectious disease field. In the cancer field, we are also seeing this specific binding data to be collected at a rising data throughput. And we're hoping that with such data volume explosion and more and more novel innovative assays being developed, such as the ones in the gray bubble, using library-on-library high-throughput screening, we will be able to collect TCR-specific training data to design even more powerful AI models to hopefully get to more accurate predictions on any therapeutics that rely on T cell immunogenicity as their mechanism of action. And internally, we're also in parallel developing our own wet lab assays to first validate, be able to validate this model performance, benchmark and be able to complement all the external landscape. So this is a very exciting time for our antigen selection, and we're very happy to leverage AI for all of these research activities. Thank you.

Thank you very much to all of our presenters. I think that was very helpful for our audience. And so I just wanted to especially thank everyone who also helped on the technical side. Thank you very much.