Absci Corporation (ABSI) Earnings Call Transcript & Summary

Good morning, everybody. Welcome to the 42nd Annual JPMorgan Healthcare Conference. My name is Dan Delfico. I'm one of the associates on the healthcare banking team here. This morning, it's my pleasure to introduce you to the Absci team. With us today, we have Sean McClain, Founder and CEO. Just a reminder that at the end of the presentation, we'll have a Q&A session, and we'll be passing on to mic. So with that, I'm going to turn it over to Sean.

Great. Thank you, Dan and JPMorgan for hosting us here. So I'm Sean McClain, the Founder and CEO of Absci. We're a data-first generative AI drug creation company. And you may recall from last JPMorgan, we announced some really exciting breakthroughs in generative AI for biologic design, being able to design an antibody from scratch with a generative AI model, being able to feed in a target of interest and then being able to have a binder to that target. Now that was a big technical feat from an AI perspective. And we took it one step further this last year. We were able to show not only could we generate a binder to a target of interest, but we could actually create a differentiated drug asset with the generative AI, being able to create a potential best-in-class, and this is for TL1A, which I'll be talking about in the presentation moving forward. And what this is enabling is being able to generate better biologics faster for patients, and the momentum that this has created has been really phenomenal over the last couple of months, and we're taking this momentum into 2024. Some of the recent momentum and success that we've had are our new partnerships with pharma. We recently signed roughly $1 billion in overall value with 2 very important partners of ours, AstraZeneca as well as Almirall, as we are focused in oncology and dermatology. We additionally achieved our 10 active program guidance for 2023. And we've built out a world-class internal asset portfolio, built by a world-class team of drug hunters and AI scientists. And in that pipeline is our TL1A asset where, again, we used generative AI to design a potential best-in-class asset. Now how does this platform work? The work is for Integrated Drug Creation platform. It's data to train, AI to create and, wet lab to validate. As I said at the beginning of this presentation, we're a data-first generative AI drug creation company. We start with the data. If you look at GPT4, it was trained on the whole Internet. We do not have that much data within biology, in particular with drug discovery. So we've built scalable wet lab technologies at Absci to train these generative AI models. Now what are these assays. These assays are built on our E.coli screening platform. What we've been able to do is actually scale the production of antibody drug candidates that we want to test and screen, so we can go from producing traditionally in the industry, thousands or tens of thousands to millions. And this production then goes through our a screening assay, where we're able to interrogate every single E. coli that's producing a different drug candidate and look at the binding affinity, what epitope is it binding, the drug functionality and the manufacturability. And this data, along with publicly available data is fed into our proprietary AI models to ultimately then use the AI to create. And as I said at the beginning of this presentation, we came out last year with a breakthrough AI model, a de novo design model, where you can design an antibody from scratch. How does this work? You take a structure of a target, whether it's a known structure or something you created an [ AlphaFold ]. You feed that into the model, and you're able to specify the epitope that you want the antibody to bind to, and you're able to then create -- have the AI create the CDRs or the regions of the antibody that binds to the target from scratch. Now these models are not 100% accurate. We need wet lab technologies then to be able to validate the accuracy of these antibodies that are generated with our AI models. We're constantly always improving. The models are never going to be where we want them to be. We're always going to be adding new attributes. So having this iterative cycle where you can generate data to train and then also using that same technology to validate is super important. And we're able to validate over 3 million unique AI-generated antibody designs in a single week. And that whole cycle time from training to validation, all occurs within a 6-week time period. Now this is what's allowed us to have the success that we've had on these breakthrough de novo models is being able to rapidly iterate. We're able to figure out what data sets do we need, what is the architecture of our models. And this is, again, what's allowing us to continue to increase the accuracy of our models as well as being able to add in new attributes. And additionally, we need these AI models to actually generate differentiated assets. And again, outside of the binding affinity, we're looking at the functionality and going into in vitro and in vivo work. And again, this is all done at our facilities using that core technology I just described on the previous slides. Now this Integrated Drug Creation platform and this AI -- these AI models are being applied throughout the drug discovery process, from novel target discovery using our reverse immunology platform to de novo design of antibodies to then the AI lead optimization of these antibodies. And what I'm going to be talking about in this case study is the last 2, using this de novo model along with the -- the AI lead optimization model to create a potential best-in-class asset with TL1A. Now what value are these models actually providing to our own portfolio and to our pharma partners? They're creating a differentiated -- they're creating biologics that are differentiated, that are faster to market and at lower cost. We're able to use these to access new biology, partnerships like with AstraZeneca and Almirall want to use these for targets that have been hard to drug, like GPCRs and ion channels. That's enabling first-in-class assets to be discovered with these models. We're able to increase overall probability of success by being able to hone in on all the attributes you want the first time. So being able to use these in a multi-parametric way, honing in on the potency, the affinity, functionality, manufacturability, the first go around. This is enabling and building in best-in-class features and enabling you to have best-in-class assets. Additionally, we're able to reduce the amount of time it takes to get into the clinic. As I said at the beginning of this presentation, we have applied this technology and actually shown that this is possible, not only to create a best-in-class asset, but to actually reduce the time to generate the TL1A asset that we've generated with this model. We did it in -- to get the in vivo efficacy, we did it in roughly 14 months, and we'll be in the clinic or have an IND in roughly 24 months. So we're going from essentially 5.5 years to get a drug into the clinic in roughly now 24 months or 2 years. And that's costing us $14 million to $16 million versus the $30 million to $40 million to $50 million normally it takes to get an asset into the clinic. And the last aspect that I'll talk about, which I actually find quite interesting is on the IP side. So since we're able now to search a much larger search space with our models, we can now actually start to differentiate on an IP level. But in order to actually get claims issued, you have to enable it. So it goes back to actually the wet lab validation. So we're able to have the models generate a lot of sequence diversity that you can't see, and then we're able to enable those patent claims by actually going into the wet lab and validating those. We can validate over 3 million unique AI-generated designs in that week time period I have just talked about. And that's enhancing IP protection in a way that we haven't seen before. So it's creating differentiated assets that have IP protection that, again, hasn't really been seen without the use of generative AI. Now the platform has been validated by industry-leading partnerships. The last 2 that we've recently announced over the last couple of months are AstraZeneca and Almirall, focusing on oncology and dermatology. And these are roughly $900 million in overall value, and that's in addition to the royalties. So we're able to share in the upside with our partners if the drug is successful. Now on the right-hand side here is an illustration of how our partnerships are structured. So we have an upfront payment. And then we additionally are paid for the research funding. And that's in addition to the upfront, that covers our costs. And then if a partner likes the asset that we've developed for them, they can in-license it, that's a milestone payment. And we continue those milestone payments throughout the clinical development process. And then ultimately, once the drug is approved, we're able to share in the upside with royalties. Now AZ and Almirall are not the only partnerships and collaborations. We have over 16 industry-leading active programs that are currently being worked on at Absci along with 3 internal pipeline programs. I'd like to see -- one thing I really love about this industry is it's a team sport. In order to get these better drugs to patients faster, we all have our own domain expertises and partnering with leading institutes like large pharma companies, like Merck and AZ, they bring that disease biology expertise, the late-stage clinical development, the manufacturing. We're bringing our AI models. And you're able to then ultimately, again, with these partnerships, get better drugs to patients faster. But we're not only partnering on the drug discovery and development side, but we're also partnering on the data and compute side, which you can see on the left. We have a leading partnership with NVIDIA that we signed a couple of years ago. And then additionally, we have data partnerships. So not only are we generating data in our own wet lab, but we have data partnerships to feed into our reverse immunology platform to discover new novel targets. Now let's dive into the internal pipeline that we recently disclosed at our R&D Day back in October. We have 3 internal programs. 2 of them are potential best-in-class. The other is a first-in-class. The first 1 is our TL1A asset, which I'll be diving into further. The second is a dermatology program, ABS-201. And the last is an immuno-oncology first-in-class target, which came from our reverse immunology platform. I won't spend too much time on this slide. This talks about the clinically-validated TL1A asset, and it's a large market opportunity, in particular, for inflammatory bowel disease, but there's also some really other exciting indications that are out there outside of IBD. So in order to validate that generative AI can actually create better biologics faster, we apply this to our -- we applied this to the TL1A asset. This is the first asset that we have applied the de novo AI model towards, where we were able to take the TL1A structure, feed that into the model and then -- and specify the epitope we wanted the antibody to bind to. And the model was able to generate CDR sequences that were unique that bound to the epitope of interest. Again, this is the first time we've implemented this technology. And all of this work that I'm going to be showing you was done in 14 months, in vivo validation of this antibody using this technology in 14 months. This is really a breakthrough in terms of how quickly you could generate a potential best-in-class asset. So we're not just looking to show that we can design a binder. We want to design this to have differentiated properties. And the properties that we looked at to really be differentiated. We're looking at, can we engineer this to have higher potency, have longer dosing intervals, so extended half-life, do subcu injections versus IV and also have favorable developability. And then additionally, I think this is a great use case from an IP standpoint, how could we use the AI to get outside of the existing IP from the competitors, but then carve out new IP landscape for our own, which we have successfully done with this program. Now let's dive into the data here. So we started out with the model generating all these unique antibody sequences. We went in the lab. We tested them. We narrowed it down to 3 advanced leads ABS-101-A, B and C. And if we look at the affinity and how it compares to the two innovator molecules, which are designated as MK-7240 as well as RVT-3101, the Merck and the Roivant competitor molecules. You can see on the left-hand side that we are able to generate antibodies that have superior or equivalent affinity to the competitor molecules. Now how does this then translate into potency? It has favorable potency when compared to the innovator molecule. We did actually show that we could have superior potency when compared to the innovator molecules. And this potency asset was an apoptosis inhibition assay in TF-1 cells. And again, this is showing that we could generate an antibody from scratch that had higher potency than the 2 innovator molecules that are currently in the clinic. Now one of the other aspects that we wanted to engineer was increasing the overall half-life so we could have improved dosing intervals. So being able to potentially go from once a month to once quarterly. So we did Fc engineering. And on the left-hand side here, we have the in vitro assays, where we showed that we could increase the Fc or in recycling, and indeed, that is the case. We then wanted to show in vivo that this is truly the case that you could have increased half-life. And you have preliminary data on the right-hand side that shows that ABS-101-B does have improved half-life compared to the Roivant competitor molecule, and the rest of the in vivo PK studies here will be done in the next couple of weeks. But again, this shows that we are able to engineer this molecule to have better half-life leading to hopefully improved dosing intervals. Now the last piece of data I'm going to share on this asset is the epitope mapping. So as I mentioned at the beginning of this presentation, we can use the AI to condition on the epitope. And one of the interesting things about conditioning on the epitope is you can change biology and you can differentiate on the IP front. So what's interesting is that the Roivant molecule is known to have a high ADA. It doesn't have a great immunogenicity profile. But the Merck molecule actually has very favorable ADA response as well as leading to better -- lower immunogenicity. And we believe that this is actually due to the epitope. So what we did was we conditioned our model on the Merck epitope. And you can see some of these actually are in slightly different or there's a little bit of an epitope drift there. And we wanted to do this one from a biology standpoint, but then two also from an IP standpoint. And so we're able to achieve a superior potency than both these molecules, but hit a different epitope, which we believe will lead to actually lower immunogenicity in the clinic. So again, being able to use generative AI to differentiate on biology, but also differentiate from an IP perspective. Now what does this look like altogether in one package? How can we use this to create a best-in-class profile or an optimal therapeutic profile? Well, if we look at where we want to be with this asset and the preclinical data that we have does support us being able to achieve these attributes, we will have a potential best-in-class here. And so we do believe we will have lower immunogenicity due to targeting the same epitope as the Merck molecule, higher bioavailability, being able to do subcu auto injection and going from once quarterly -- once monthly dosing to once quarterly. And again, we believe that this will create a best-in-class TL1A asset, all designed with generative AI. Now going on to the time lines. So it took us 14 months to generate a differentiated profile that was in vivo validated. And it's going to take us roughly another 9 to 12 months to get our IND enabling studies done and be in the clinic or have an IND in roughly 24 months. And so we're going to be initiating our IND-enabling studies in February with the goal to have an IND in Q1 and then enter into our Phase I shortly thereafter. Again, I just want to highlight this. We're using the de novo platform to be able to create a differentiated asset and be able to do it in a very, very rapid time frame. This is really the future of how drug discovery is going to be impacted by generative AI. Now it's not just AI that is helping move things forward, it's a mix between the human consciousness as well as our artificial intelligence. And the team that we have here is extraordinary pioneers in the field. I mean we have world-class drug hunters, like Andreas Busch and Christian Stegmann along with the rest of the team that have [indiscernible]. We have world-class AI scientists at Absci, world-class synthetic biologists, all coming together to ultimately see this vision through and accomplish what we've accomplished to date. We additionally have a state-of-the-art 77,000 square foot campus, where all of the data is being generated and all this validation work is being done because, again, the success that we've had is not in AI-first approach, it's a data-first approach. It's that integration that allows us to do what we do. And we've recently, over the past 2 years, built out an extraordinary Executive team and Board. And one of the recent Board members we brought on, which we announced yesterday, is Sir Mene Pangalos. He is the Executive Vice President of Research at AstraZeneca. And it's just a testament to bring on somebody of Mene's caliber to really validate what we're doing here, and it's really an honor to work along with many but my executive team as well as the Board. Now in summary, going back to the beginning. Last year, we showed that we could use a de novo model to create an antibody binder, but now we've taken it one step further, where we're actually showing preclinical data, where we can differentiate on the biology, creating a potential best-in-class, being able to create biologics faster than we have previously at lower costs. And this is all happening because of our Integrated Drug Creation platform, our data to train, AI to create and wet lab to validate, being able to do that in a rapid 6-week time period, being able to validate over 3 million unique AI-generated designs in a week's time period. And the platform has been recently validated by industry-leading partnerships, like AstraZeneca and Almirall. And we've shown -- actually shown that generative AI can create a potential best-in-class asset with TL1A. Now this momentum that we've had come in into this year for the momentum we had at the end of last year is going to continue into this coming year. We have exciting catalysts coming up in 2024. Not only are we going to continue to sign breakthrough partnerships, like the ones we've signed with AZ and Almirall, but we're also going to have exciting data readouts. We're going to be entering in our IND-enabling studies in February for TL1A. We have some exciting in vivo validation studies that will be completed for ABS-301. That's our first-in-class IO target coming from our reverse immunology platform. And we're going to have a development candidate selection for ABS-201, which is our best -- potential best-in-class dermatology target. And then we'll have the IND submission for ABS-101 in Q1 '25. It's a really exciting time to be in the field of generative AI and biology. And we have now taken what was once a vision and really put concrete data behind it to show that this actually can make a difference in making better biologics for patients. Thank you.

We'll open it up to Q&A now, and we'll have mic coming around. We'll also have Zach Jonasson, CFO; and Andreas Busch, Chief Innovation Officer, coming up for the Q&A portion.

Great presentation. Is there a reason to -- like how many paying pharma collaborations do you think you could do? Like is there a reason to -- like is there a max amount that were above that becomes a focus issue? Or is it better to have 5 or 10 or like what does great look like on that pharma pipeline?

Yes, absolutely. We really want to be able to use these pharma partnerships to go after disease areas that we don't have expertise. And so it allows us to create this differentiated portfolio of assets that we couldn't do ourselves. Like I think oncology is a perfect example. Like we wouldn't plan at this point in time to take something into late-stage clinical development in oncology because we don't have that expertise. But who has that expertise is, is AstraZeneca. And so we see this as helping create differentiation for our portfolio and then using that upfront cash then to focus in on the assets that we find really exciting for our own internal pipeline to continue to develop those. And so we see this as a kind of -- we're going to continue with these partnerships, but they're going to be very focused. And I think we're going to see them be more focused and much deeper in terms of how integrated we want to get with some of these large pharmas because they do bring expertise that we don't have, and I think we can have that synergy. And so we do see it continuing to grow, but I don't think you're going to see it expand, I guess, depth-wise in a number of companies, but you could see it expand depth-wise on the number of programs per large pharma partner.

I'll just comment, too, Charlie. One of the things that's been exciting over the last year is looking at the efficiency gains in that cycle that Sean described, where we generate data, we train the models, we validate. So we're seeing efficiency gains, and we're seeing the models improve. So that gets to more and more capacity as we look into the future.

Thank you for the nice presentation. And you talked about Immunogenicities related to the epitopes. I'm curious about the reason of the statement.

Yes, absolutely. So one of the reasons we believe that it is that the epitope -- because if you look at like the T cell activation assays, the profile for all 3 actually looks pretty favorable. And so -- we think it's more of a complex issue, like stimulating the B cell response versus more of a sequence liability. So we think it's more structural-based immunogenicity of where it's actually binding than a sequence liability. That's at least our hypothesis at the beginning. Now we'll have to see how that translates into the clinic. But that's kind of the best scientific hypothesis we can come up with at the current moment.

Maybe to bring this back to the internal pipeline and some of the partnerships. How does the growth of the internal pipeline changed the business strategy, if at all? And from there, how have you thought about maybe potentially partnering some of the internal assets?

Yes. I can take it at high level and have Zach dive a bit deeper. So one of the things that we see is really exciting about this platform is, again, not only creating differentiated assets faster, but it's actually enabling a new business model, where we can actually take assets further and get them to inflection points that are very meaningful in a much shorter and at a cheaper cost. And so we do plan on continuing to invest heavily in our own internal pipeline just like we have with ABS-101 and take it anywhere from a drug candidate phase to a Phase II and then sell it off at that point in time. And I think like what you're seeing is that pharma is reaching earlier and earlier into these partnerships. If you look at like I think the Harpoon-Merck acquisition. I mean, that was a Phase I oncology asset. They bought for $600 million. And if you can, in a very rapid manner and in a very cost-effective way, get to a Phase I with an oncology asset, let's say, for the 301, that could be -- these economics start to make a lot of sense. So instead of partnering at the target phase, where you're waiting for these milestone payments out in the future, you can actually bring these payments forward, bring cash in and kind of repeat that cycle. Again, we don't plan on taking anything late stage, but I do think it enables a new business model that is able to kind of bring in these payments a lot sooner with higher upside because we're taking them to differentiated value inflection points.