Merck & Co., Inc. (MRK) Earnings Call Transcript & Summary

Welcome, everybody, to my favorite holiday of the year, Ginkgo Ferment. I am so excited for this absolutely -- am I allowed to say? Badass panel of women talking about AI. I promised to the panel that I am not going to ask or talk about anything that you can find on Google. We want to take this a little bit deeper than panels like this normally go. All of these women have had incredible careers, sometimes a second or third careers, in AI, have been focused on this space in decades. And we're really excited to have a conversation here.

I wanted to start with the opportunity in AI. You all have decided to spend your lives focusing on AI. It's a buzzy topic, but you get up every day, excited to work on this problem. And I want to know what is that thing that you are most excited about transforming in the next few years, the next decade of your careers? Maybe, Karen, I'll start with you.

Yes. Great. It's great to be here. So one of the things that I think we have an opportunity to address right now is the fact that we know so little about human biology. In all of history, we have essentially tested and developed and launched about 1,300 drugs when there are about 10,000 diseases. So we have a long way to go. I think the other thing is when we think about chemistry, we've essentially as an industry only tested enough compounds to fit in a bucket of water compared to the entire oceans on the earth. So that tells you that there's so much headroom that we haven't yet explored that helps us, I think, as a society, go ahead and try and solve for human disease.

Molly?

Awesome. Well, thank you for having me, I'm really excited to be here. So I feel like you've gotten to a place over the last 5, 6 years where this vision of being able to, first, represent biology in AI, and then engineer it is becoming a reality. And that wasn't true 5, 10 years ago to the degree it is today. And so the questions that I'm thinking about now, where we go and were we take that, are really about how do you do science? And how do you think about science as a scientific endeavor? And ask questions, ask -- test hypotheses and really start to expand knowledge and add to the world that we don't -- that to your point, we know actually very little about how to expand knowledge. And the types of technologies that are being developed and the types of things Jason talked about today are the keys to being able to do that, and there's more to do. We know today that LLMs, for example, are very bad at using numbers. Science is a quantitative endeavor. So how do you start to integrate? How do you generate the right numbers to give us the right answers, and generating hypothesis and test hypotheses and update models? So those are the types of things that I'm thinking about that aren't necessarily super near term. Like I'm not saying this is easy, I think it will be hard. But that's what keeps me going.

Iya?

Is it working?

You have speak up.

Yes. Okay. Perfect. All right. Thanks for having me here. Great conference. I really enjoyed the keynote, right, talking about automation and scale. So I want to just kind of work backwards. I really feel like we're living at a time where a lot of what was either incurable or difficult to treat from a disease modification perspective is actually now like we're at the cusp of being able to do that, right? We can kind of see the hints of it coming from looking just at even oncology, new immunological drugs, that if you give it early enough in enough patients with combinations, you can actually cure or even change the course of the disease. And we're starting to take that thinking and apply it to other areas, like immunology. Like just take for example IBD, irritable bowel syndrome. It's a disease that, at first for patients, feels like an annoyance, but eventually it becomes super debilitating. And for me, it's what can we set to learn about the mechanisms of disease so that we can treat patients in a just much better way and find life-saving medicines much faster? And the opportunity that AI now affords today, after it's shown that it could actually do something really remarkable, right? So we can -- AlphaFold, that announcement by the DeepMind Google team, showed that if you put enough of the right data into an AI system; and leverage architectures, whether it's transformer architectures, reinforcement learning; and combine it with genetics and physics-based methods, that you can actually predict every single structure for every single protein in the human genome. AI did this in 2021, right? So now when I think about how we're going to get AI to get to that next transformation, for me, it's that level of being able to now measure what we need to measure in biology so that we can truly learn biology. And so while AI is getting better at learning, we are getting better at measuring. Soon, it will be possible to take any slice of tissue and measure every single transcript for every single cell type in every single organ in the human body. It's amazing. We finally have the tools and the technologies to measure biology at the right level of resolution, your genes, your proteins, your metabolites, every single cell, and that's scale. All right. How are you going to put that to use? How are you going to put it to use to understand biology and accelerate our ability to get to better medicines? That's what I'm super excited about.

Debbie, you want to close this off?

So I'm going to go very zoom out and I'm pretty sure you could find this on Google, apologies. But I wake up in the morning and what I want to do is build and develop models that make biology a quantitative, predictive science. And I think that the generative models that we've started to talk about now have really opened the well for that. I don't think it's the end, it's the beginning. And one of the things, again zoom out, I'd like to say, is that just as math enabled -- mathematics, as it's called here, physics, AI will enable a quantitative biology, biology for the 21st century as we expected. Now my vision, basically, after listening to Jason, he'd have already done it and I'm not going to repeat that. That's partly a joke, but I do think we're moving extremely fast in what we can do. And the vision here is making biology quantitative and predictive, and the different models and different data we'll need for that will affect disease prevention to early diagnosis, as well as treatment and monitoring and sustainability efforts. And I know we want to focus here today on disease and the prospects for that, but it's important to remember that it really is going to open up all those other fields, which will then impact treatment and democratize it.

When we were chatting earlier this week about this panel, we talked about what are the components, what are the ingredients that are needed to actually achieve this vision? And largely our industry talks about 3 ingredients. You have compute, which our friends at NVIDIA and Google and Amazon are working on. You have technology, which is getting rapidly, rapidly democratized, and we're all benefiting from that. And then there's data. And I think we all agree that one of the biggest limiters, and Iya, you were talking about this earlier, historically, has been access to data. And one of the things we're all, I think, really excited about are new technologies, whether it's the kind of scale automation that Ginkgo is building, or new machines that are allowing us to get greater resolution on biology. And so I'd love to go to a few of you and talk about where you see the greatest needs for data that are going to unlock the next real answers in biology and help us there. And so maybe Iya, because you kicked off that topic, I'll start with you.

Sure. So I think for me, the greatest need for data is going to -- so let me just give a landscape of where the data landscape really is right now, and then we can get into what we need. Okay. So right now, if I wanted access to whole genome sequences and large populations of patients that exists today. It does. There's lots of consortia out there. One of them is one we just joined, Merck joined, it's called National Biosystems. It will have very soon a mass 250,000 whole genome sequencings, along with clinical records and over time. That's amazing. There are companies and clinical centers that will give you access to tumor tissue bank, and you can get RNA and genetics and longitudinal records. And believe me, we're going to make hay of this data. We're going to find insights from that data, combined with AI and machine learning, to really understand disease better. But where we need to go now is that next level of resolution, right? Because when I take this data and I ask it the question, are there variants that can cause disease biology, I'm going to find them. But what do those variants mean? What gene does it map to? What does that gene actually do? What is its function? How is it expressed? Which cell types is it expressed in? So we've gotten to sort of level 1, all these consortia and databases, we need now level 2, we need to actually resolve what's happening in every cell, every tissue and over time and over the course of disease. And new technologies are coming on board that are going to help us do that. Technologies like -- we know about RNA, but now it's spatial, right? We can do like spatial transcriptomics and measure what's happening. And what I'm hoping folks get inspired by is, I think in about a year, we'll be able to do this whole genome, which is pretty amazing, right? Every transcript and every cell, every human tissue, we're going to be able to do it. But then I need scale. I need to be able to do it under lots of conditions, lots of perturbations. And so if the theme of Ferment is scale, then what we need to do is take some of these really cool advances in biology and measurement and scale them now so that we can see enough of what's happening in cells, bodies, tissues over time in different diseases so the AI can truly fundamentally learn biology.

Maybe Karen, because you sit at the intersection of biology and physics, chemistry and physics, I'd be really curious for your take on this question as well and where you also see the opportunities for cross-learning between these different disciplines.

Yes, sure. I mean I think you opened the panel with a question about what is the potential impact and how have things changed. If you look back in history, computers weren't fast enough. 21 years ago, maybe 22 years ago, we didn't really have human genome. We now have those things. But are we making progress towards what I think we can define as really the atomistic level of what's going on inside our bodies, right? Chemistry is really an explanation for biology. Biology is really about chemistry, and chemistry is really about physics. So I think context is everything here because when we talk about the genome, when we talk about the protium, when we talk about proteins, what we're really talking about is how atoms interact with each other at the end of the day, whether they're healthy in the context of a healthy protein or a disease protein. And I think what's exciting is that we're now at the point where we can actually start to interact with that physics. As you know, it was really hard to simulate physics in the setting of a biological system. I think we're now approaching the time, because of the speed of compute, our ability to predict these atomistic interactions, we're now at a time where we can actually reduce to practice the question of what's happening at the level of physics? And that's really exciting. In fact, I think some of you may have come across this term, that physics is actually the new data, right? We are now at a point where the sort of macro experiments we used to do, we can actually now go ahead and, at least when you're designing a drug, really focus on what's going on at the physics level. And that, I think, is super exciting. It's important to note, though, that when we talk about training sets, those training sets really, I think, are most powerful when they're most sort of accurately representing the system that we want to model. And again, I think that's where we are now at a point where those training sets can actually be developed based on the physics at enormous scale, right? We used to screen like 1 million-compound library, and we were super excited about that when I first started in the pharma industry. Now we're screening like 1 billion compounds in a few days. So having access to the physics and being able to scale that physics, I think -- and use it as a training set, I think, is super exciting.

And one of the things we were talking about was the ability to use these models to also inform where to generate data that is most useful to help identify the context, help identify the questions, so that you're generating data that really answers the question you want to answer. And I know, Debbie, you've got strong feelings about this. And our ability to take advantage of the data that we already have and to ask it better questions and organize it in better ways. And you're a leading expert in driving absolutely remarkable results with quite scrappy resources.

Thank you. I think -- yes.

And so I'm curious, where do you see the biggest bang for our buck in new data sets that are available?

So when -- I'm asked quite a lot about what data we would want ideally, and as I'm sure other panelists have expressed. And my question back is still, "For what?" I think to a large extent, it's okay still to be quite task-oriented because there are, despite, the fact we're going to be able to take every human tissue at every stage in every cell type, there are infinite measurements we could make or develop measurements to do. And the trick is going to be able to build those models that can learn what kind of multi-modal data, what is the minimum to do the rigorous work? I don't mean minimum like get away with it. I mean, minimum comfortably. What modalities do we need for what task? And it turns out you can learn a heck of a lot, for instance, about the thermostability from natural sequences. And we've all seen that and we talked about it even earlier this morning. And I think we've even done some studies to show you've got a couple of data points and you got natural sequences and you can pretty much predict, to some extent, the thermostability of many other proteins. Now you take another example where you want to engineer a protein which it hasn't quite done naturally, and it's got to exist in a particular cell type and report, because it's a biosensor as well, on a particular state. And that will need different kinds and different extent of the data collection. So for me, it's about building those models that can be multimodal. And then those models also learning how to learn what data to generate. So a bit circular, but...

I appreciate that. So there's data and then there's an organizational question that I think we talked about. And I think that organizational aspect has to -- comes into play within companies and the sequential nature of the research process today, especially in the large pharma drug development cycle. But also, I think there's an interesting question about the role of the academic and open science community versus the role of private industry. And so maybe, Molly, I'll go to you because you're in a really interesting spot in the ecosystem where you're incubating new ideas and you're really sort of sitting in the middle of this. And I'm curious, organizationally, how do you think we most quickly accelerate our learning and our ability to bring patients new medicines and bring new materials to the world on these domains?

Yes, definitely. So when I hear this question about how do we organizationally transform, how we bring [ sophistications ], the first thing that comes to mind is the people and the culture and what we're actually building. I think one of the things that is really true about this industry and about what we're doing today is that people have to be brought into this future of a connection between technology and biology. And you've seen this throughout history that some people just don't buy in. So how do you build a culture where people are bought in and they see how their world is changing? And I mean, it's drawn to the image that Jason put up this morning of interacting with a computer 50 years ago to how we interact with this -- today and how our kids are going to interact with it in the future. I see that transition of technology happening even more rapidly in the next 20 years than it has in the last 50 years. And so we're all going to have to rethink how do we build organizations? How do we interact as a community of people, either in industry or an academia? And how do we connect those communities in the right ways? And a lot of it's about how do we find the right research questions and then how do we apply those to the most important problems that we face today? and bring in those together in a way that they're synergistic, not competitive. And that's a lot of what I spend my time doing, is thinking about how do we identify the unique insights that come from many people? It's not usually one unique insight. It's not one person. It's not -- it doesn't -- at least in my experience, it doesn't felt like a genius in the center. It's felt like a community of ideas coming together to have a greater whole. And so I think that's how I spend my time thinking about this. So I interact a lot with academia as a part of this, bringing in many ideas to create something bigger together, and there's a role for everybody in that. And it's about the culture, the people, the community that is going to transform science in the future.

Maybe Iya, I'd love to get your take on this as the Head of AI in a large pharmaceutical company. How are you seeing the industry react to the emergence of AI, the importance of AI? Everyone now is trying to adopt it, but I think in different ways and at different paces. And where do you see the roadblocks? And how do you think the industry needs to evolve that R&D process?

Yes. So I first of all, I think every pharma company is going to have its own strategy around this. And a big part of it is, as many of you know, to get -- to go from like an early discovery to drug to market, it's not what can AI do, it's what can AI do for target discovery? What can AI do for a small molecule discovery? What can AI do for biologic discovery? What can AI do for predicting safety outcomes? And what can it do for identifying the right patient for the right drug at the right time? So it's many different levels of AI. And different pharma companies have organized differently to go from what I described end-to-end. But I'll tell you how we're thinking about it, right? So the key for us is you have to embed, that's it. So your small molecule department, medicinal chemistry department has at the table folks running the large assay screens, it has a physicist maybe even thinking about physics-based models and how to integrate that with transformer-based models, and it's got your AI architect engineer. And it's all 1 function, 1 unit advancing our ability to develop small molecules faster. So you got to embed those folks at every single function end to end because your goal isn't just to find a great target, a great drug, a safer drug, you got to do all of that end to end. And so the big thing is, how do you track the right talent? And how do you get folks engaged and wanting to do this? And wanting to this not just like as a kind of little tiny project, but excited by the scale of what's possible in pharma and as well as even how pharma interacts with the biotech system. So we take that very seriously. And we've set up initiatives that are both around getting the right talent inside of pharma and inside our departments. But as well as once they're in, that they're part of communities of practice, communities of experts, and they can really move the needle within each of the functions that they serve.

Maybe Debbie to wrap out this topic. How can industry better partner with academia?

Well, I think it's a really interesting question. I think when I first started my lab, which was about 9 years ago, it was actually about what are you encouraging your students to do? Where are they going to go and do their post-docs? And what faculty position? They better go to this conference. And I said, "Well, hang on a minute." And they were saying, "Don't let them go to industry. That's the bad ones." They were still saying that. And we all know that, that's fictitious. And we know that it might have -- even if it ever was true, it's certainly not been true in the last 10 years. And what's more, I think we've seen -- now I'm going to put down academia. And we've really seen huge development in AI methods in industry that have then come back into academia. And I think I really want to emphasize this point about the fluidity that we need. And I think there is fluidity, people going back and forth between industry and academia. And I think we'll see it a lot more. And at the moment, it's very serendipity and it will be really good to have mechanisms in place where that's organized and encouraged and students are empowered to do that, and we can train together. And I think that's going to be -- mechanisms for training together is going to be a big step that we need to take.

I know that we have a diverse audience. We have folks in industry, we have many students here. I'm sort of heartbroken, we're just about out of time, and I'd love to talk to you all day. But I am hopeful that, if nothing else, this panel raises the bar a little bit for the industry, and that none of us are ever again the only woman on a panel, the only woman at a dinner on AI. And I'd be curious, just for the folks that are in the audience. If you have one piece of advice to folks that want to build a career in this field, you may not have a lot of role models, what would you say? Karen, I'll start with you, and we'll go down.

So I'm going to borrow something from linguistics, right? I think being bilingual or trilingual is going to be really important. Actually, we're spending a lot of time...

Do you mean actual language, or do you mean like...

What I mean by that is -- and it comes back to what I said. Our teams are really interesting. We've got physicists, machine learning experts, medicinal chemists, synthetic chemists, our CMO, right, all sitting together, thinking about how do we solve problems? What we're seeing coming up through college, through even high school now, are people who don't train in silos. They actually are bilingual. They know something about data, they know how to handle data. But they also have a sense of medicine, right? So these people who are bi or even trilingual, medicinal chemists who know how to write command code, this is, I think, a super exciting opportunity for the next generation. And so I'd say don't train in a silo, be bilingual, trilingual.

Thanks. Molly?

So I think one of the big things to keep in mind also in addition to -- I totally agree with the bilingual in nature, is to really find your superpower and believe in yourself. I think there's a lot of trying to fit into what exists. And I think the more of us that are identifying the ways to create new opportunities and believe in yourself and believe in your vision is a part of that. And community and allies, men and women, all backgrounds in doing that. Just believe in yourself.

Iya?

Awesome. This, all of it, great advice. So I can only add to what's been said. And it's -- so 20 years ago, when I started in this field, I was always back to what you've had said, the only woman. Whether it's at the panel, the room, the this the that. I think it's changed a lot. I hired my Head of AI/ML last year, she's a woman. Now I wasn't looking for a woman. It just so happened that's the resume that stood out. But I think it's because, given that we're like in this new field, right, and when something is new, you have a real opportunity to kind of join it and shape it. And so my advice is, it's a new era, don't go by what has been. Really dig into where we are today and realize you have the opportunity to be a part of it, to shape it as part of the ecosystem, above men and women. And then lastly, just one word of advice, this is about targeting all the women out there: Just be fearless. You're not going to fear, you can do this.

Debbie, take us home.

All right so all of the above, and especially the last comment. I was asked this about 5 years ago, I was on a panel for the Chan Zuckerberg Initiative. And I said -- they said, "What's your advise to everybody?" I said, "Go big. Go bold. Learn statistics." And now I would add one more thing to that, actually is go big, go bold, communicate, talk to people. And learn statistics.

All right. Learn statistics. Thank you all. Enjoy Ferment.