Absci Corporation (ABSI) Earnings Call Transcript & Summary

I'm Bartosz Ostenda, I'm a partner within investment banking. I run our Tools and Diagnostics business, and it's my pleasure to introduce you to Sean McClain, who's the CEO and Founder of Absci; and Zach Jonasson, who is the CFO and ex-Board member of the company.

So with that, let me maybe start by asking a question about the strategic journey of Absci in the pivot of the years towards AI drug discovery?

Yes, absolutely. Thank you so much for having us here. Bartosz, as you pointed out, we originally were not an AI drug discovery company. We started off building a technology that could scale protein-protein interactions. We figured out how to engineer E. coli to produce antibodies and develop the screening assay that was able to interrogate every E. coli that was making a different antibody and figure out how it bound to that particular target of interest, the epitope, the overall affinity. And this was right around the time that transformers were taken off 2018 with Google, the first generative models. And it was this idea of like if you could take this data, this protein-protein interaction data, with these generative AI models, you could really go from this paradigm of searching for a needle in the haystack to actually creating it. And what this is enabling is us to be able to go after these hard-to-drug targets, like GPCRs and ion channels. And that's what partners like Merck and AZ and Almirall have gotten really excited about and why they decided to partner with us is because we're actually able to leverage AI to unlock new biology that existing technologies aren't able to ultimately address. And so we see this generative AI platform as, yes, 1, unlocking new biology, but then also being able to drive the time and cost it takes to get therapeutics into the clinic down significantly.

Can you give us a high-level overview of your AI platform and how it differentiates to other AI biotech companies?

Yes, absolutely. So we're really focused on the design. And so how the model works is you take a target of interest, whether that's derived through AlphaFold or the cryoEMs. You take the structure, you feed that into the model. And then you condition the model on the epitope of interest, and the model then is able to then design the CDRs of interest that bind to that particular epitope of interest. And this is the first time anybody has really shown that you can get epitope specificity with any sort of approach before. And again, kind of going back to why is pharma interested in this? Why are we excited? Is this ability to do epitope landscaping, be able to, let's say, take a brand-new novel target that you don't know the biology on and be able to generate antibodies that could hit all the surface-exposed epitopes and take that into the wet lab and figure out, okay, in cell-based screening -- in cell-based assays, which of these antibodies gives me the biology I'm looking for. And so you're able to unlock new biology and get the biology faster than really ever before. And again, it goes back to being able to design antibodies with the attributes you want the first go around, getting the epitope of interest, the affinity, the developability. And again, we see this as being able to unlock new biology, in particular with ion channels and GPCRs.

So you mentioned something interesting, which is you're going to design a molecule, you're going to use the in-silico methods, then you go back to the wet lab, that's everything. Where do you think we are on that path of being all wet lab to being wet lab and in-silico, is it ever going to be all in-silico and what do you think the future is holding for us in this regard?

I think ultimately, yes, you will be all in-silico, but I would say that the reason we've had the success that we've had is because we've had that lab in the loop. We've developed this technology that is generating data to train these models, but not only to train them, but to validate them, and we can validate over 3 million unique AI-generated designs in a 6-week time period. And that's an important cycle time for us being able to do an active learning loop in biology in 6 weeks is incredibly fast, and that's what's ultimately allowing you to improve the overall accuracy of these models. And we're going to get to the point in the future where the accuracy of this de novo model is you no longer need to actually go and validate 1,000 antibody designs. You can go validate 1 or 2 designs and then move forward with a drug candidate from there. And then ultimately, once you kind of solve that design problem, you then have to then go on to the next problem and the next problem is biology because design isn't solving biology. It's allowing you to test hypotheses that you've never been able to test before, but it's not predicting biology. So I think the next phase in generative AI evolution is how can we use AI to actually predict the biology where you have a brand-new novel target and for a particular disease and you can say, "Oh, this is the epitope that's going to give me the biology that I want." But we need new technologies, new wet lab technologies in order to get there, and we don't have that currently. And so I think you're going to solve 1 problem, and you're going to be in-silico with that, and then you're going to go onto the next problem, figure out how do you scale the data to solve that problem, you'll ultimately go in-silico with that and then you go onto the next. And so you're always going to continue to have this lab in the loop process, and I don't see it going away anytime soon.

So what do you think is, at this point, the biggest challenge for AI drug discovery companies, both from science and computational power perspective or algorithms or otherwise?

I mean I would come back to Sean's comment about lab in the loop. And like our focus over the near midterm is to make that lab in the loop process more efficient. As Sean mentioned it 6 weeks today, 1/3 of that time is just getting the DNA. So we're working on making that more and more efficient. And so when you look across the landscape, if you don't have that capability to generate useful data, you really can't train models. And if you don't have the ability to go validate in the lab at scale, you can't improve or test those models. So I think that's the key piece of this is having the right training data and the capability to do the validation and then run that iterative loop. So we've been doing that for 4 years. And so I think I'm excited to see where our model performance is today, but even more excited to see where it is tomorrow because we're not on a linear path. The more data you generate, the more validation you do, you are on an exponential learning curve. I think it's very exciting. So for other companies that don't -- and I used to be on the investment side, investing in this space. And there's a lot of companies that use AI in different ways, but there are very few that actually have a data capability, let alone an integrated lab in the loop process that's really functional at high scale.

So tell us a little bit more about your data, how differentiated it is? How it contributes to your success? Where is really -- what is the detail that allows you to deliver better performance than anyone else?

Absolutely. So we take a multimodal approach to data. So we use both publicly available data like the structures that are publicly available. We use that in the model. But then we also generate our own proprietary data, which is functionally based driven data and sequence-based data. And this data is how an antibody interacts with the target? What epitope is it binding to? And what affinity is it binding to? So you have this sequence structure and function-based model that, again, is taking in different types of data to ultimately get it to the point where you can actually de novo design molecules. I know roughly 2 years ago, we came out with a publication that showed indeed, you could take both the proprietary data we're generating plus locally available and generate this model where you could take a structure of a target, condition it on an epitope and have the model generate the CDRs that would bind to that particular epitope of interest. And the model is continually improving in accuracy because we have this active learning loop, that 6-week time period where we can generate new data for the model and go and validate it. And that increases overall accuracy. So it tells us what data we need and then also we can rapidly iterate on new model designs and architectures as well. And ultimately at the end of the day, models are going to get commoditized. But ultimately, the moat that you're going to have is going to be around the data. And so that proprietary data that we are generating is hugely advantageous not only for the model working, but also the competitive moat that we are generating.

I think that goes well with my next question, which is competition from all the large AI players who have access to infinite amount of capital, has incredible computational power with all the resources behind them, how do businesses such as yours compete effectively? And what is that competition based on in your case?

Yes. I mean, look, it comes back to having data in a wet lab that can support that active learning cycle. We have tons of respect for companies like Google and OpenAI and NVIDIA. We don't see them as competitors. We see them as potential partners. We can work with them on the compute side. But at the end of the day, without that ability to generate data and do the validation, you can't improve models. This is not a situation where we're training a chatbot like ChatGPT, where we have the entire database as the Internet to search, that doesn't exist for biological data. So it's really important to be able to create that data and then to go and validate your models at scale. So that's the key barrier that I think prevents those companies from coming into this space effectively. But again, we tend to view them more as partners and we have a partnership with NVIDIA to that point.

Yes. And like with these chatbots that are out there, these large language models, they're ultimately compute limited. But if you look at biology, as Zach was saying, the biological data that exists out there compared to everything that's on the Internet, it's a small fraction of that. And we're finding out that we are data limited, not compute limited at the current moment. And you've seen these models, these large protein language models that have increased overall compute, but the accuracy of the model doesn't increase all that much, which then goes back to showing that we are data limited. And again, the accuracy of our models wouldn't be where they're at if we didn't have our own proprietary data feeding into these models. And so that's how we compete. We compete on the wet lab side, and it's the wet lab on the data generation, but also the cycle times, that 6-week cycle time in that active learning loop, I mean normally, that would take normal companies that didn't have our technology at least 9 to 12 months just to do one iterative loop. And so by having that, that 6-week cycle time that gives us a huge advantage to create these models. And by creating these models, you're, again, unlocking new novel biology, and that's what's driving these partnerships that we have and is what is driving different -- our differentiated assets that we have in our own pipeline.

The results that you are getting and all of you having spent so much time in structural biology and proteomics, does it happen that you get results that are completely surprising to you? I would think very often, you're going to get a question and going to say, well, maybe there is going to be this kind of motif or that kind of motif, this should work, that should work, and then you feed it through your system and you get something completely unexpected?

Yes.

I was -- there's 2 ways we're surprised sometimes. One is the way you described where maybe it didn't work as expected. But the reality of that is that helps us improve the model and helps us generalize, when we find places where it's not applying to a certain type of structure. That's why we want to do more and more molecules, and that's why partnerships are important. But the other way we're surprised is by the capability of the model. And I can't give too much detail about this, but we have an active collaboration with UCLA, with Dennis Slamon Group. And what we found with our most recent version of our Gen 3 models is that not only can we design for an epitope, but we can massively like sample the interface of that epitope, design antibodies that have novel interactions, even with known epitopes that deliver not only higher potency but novel MoA. And so that is really exciting and I think pretty much unexpected by the field. So we're looking forward to publishing on that later this year.

So as you change your focus a little bit as a company and direction, how has your target business model changed? What is your target business model now that you've started to discover, develop your own compounds and you have pharma collaborations, what is the target? Trying to be like antibody companies a couple of years ago, what's the goal?

Yes. I mean really the goal for us is a hybrid business model. We've previously focused solely on partnerships. Now we're focused on both partnerships as well as our own internal pipeline because we've seen how the model can generate a lot of value for our partners, but we're not -- when we partner at the target phase, we're not able to retain as much of the upside as we think that we should based on the value that we are providing. And so we've seen that we can create really exciting differentiated assets. In addition, we're seeing that we can also drive the cost down to get to an IND, normally takes $50 million to $100 million to get 1 asset into the clinic. With TL1A, our lead asset, we were going to be in the clinic with an investment of $13 million to $15 million instead of that $50 million to $100 million. And so you're able to get a lot more bang for your buck with the investment you have. So instead of investing in 1 asset, you're able to invest in 5 to 7 assets. And additionally, you're able to get into the clinic a lot faster. So instead of 5.5 years, you're able to get into the clinic within 24 months. And so we see this ability to develop our own pipeline as being able to retain more upside and get to value inflection points faster and cheaper than traditional biotech companies would have in the past. And so we plan on partnering our own internal pipeline anywhere from IND or drug candidate phase all the way through a Phase II. We don't plan on taking anything past Phase II. We think that pharma is great at late-stage clinical development and the commercialization and marketing. And so it's still a partnership business model, it's just where we're partnering on the game. Is it the target? Is it IND? Or is it Phase II?

Absolutely we still want to do drug creation or drug discovery partnerships. We call them drug creation. To Sean's point, we're creating the needle. Those give us diversification across indication and it also gives us an ability to work on lots of new novel targets that now helps us generalize our models. So that is -- will continue to be an important component of what we do. But to Sean's point, we see a tremendous ROI in investing and developing our own assets to proof points where then we can structure a significant transaction.

So adopting your lingo, creation of drugs, what is the process for selecting the targets? How are you making those decisions and what goes into that?

Yes. So we have an AI approach and then we have a knowledge-based approach. And so the first two targets that we have in our pipeline, ABS-101 and ABS-201, those are our knowledge base, and that's coming from the experience of the team that we have, led by Andreas Busch. He was the former CSO at Shire. Prior to that, he was Head of R&D at Bayer, has over 10 drugs approved under his leadership. And we really wanted to kind of take the team's knowledge and go and find what are kind of low-hanging fruit where we believe in the biology or the biology is semi validated, let's say, either in vivo studies or clinical -- there's some interesting clinical readouts and how can we then differentiate, how can we differentiate on efficacy, how could we differentiate on potential like immunogenicity profile or differentiate on dosing and patient convenience? And then making sure, too, that the market is big enough for us to go in. And so that's kind of our best-in-class target strategy and kind of the knowledge base and then we have this reverse immunology approach that is allowing us to get novel first-in-class assets. And this is ABS-301 came from this reverse immunology approach. This is really interesting, it's based on tertiary lymphoid structure biology where there's a lot of new research that's been coming out that show that the antibody repertoires within the B cells within these tertiary lymphoid structures within tumor cells is very different than what's in peripheral blood. And if you take those antibodies from patients that have had an extraordinary immune response, and you go and find out what those antibodies are binding to, you can de-orphan and kind of discover these brand-new novel targets. And that's what we discovered with ABS-301, which is a novel I/O target. But again, the first-in-class approach isn't going to be the first assets we take through. We really wanted to take through differentiated best-in-class, where some of the biology has been proved out and you're just taking the risk on the technology itself and really proving out the technology, the design aspect of the technology works, able to create that differentiated asset without taking the biology risk. But we do have this blended approach where we can go after novel first-in-class, but also have a best-in-class approach as well.

So when you look at the novel best-in-class, do the existing molecules feed into the algorithm when you're trying to improve on them? Or do you take de novo approach in those situations?

Yes. So TL1A, we took a completely novel approach on that. So all we started with was the TL1A, the structure of the TL1A antibody, the competitor molecules were not set into the model. And we then specified what epitopes we wanted to hit. Ideally, in our case, we wanted to hit an epitope that was similar, but adjacent to the Merck epitope because we believe that that will give us lower immunogenicity but we also wanted to be able to get monomer and trimer binding, which the existing competitor molecules do not achieve. And so that's how we went about the design aspect and the model was able to then design these antibodies from scratch without any prior knowledge of the existing antibodies that were out there.

And how many backup targets do you have to account for what you talked about before, which is the biology that we can't really model?

Yes. So I mean we ended up narrowing it down, what was it like 50 to 100 antibodies that we did extensive characterization on the affinities, looking at potency, developability and manufacturability. And then ended up narrowing those down to a handful and then we ended up selecting the drug candidate to move forward with and we have a backup candidate as well. And so you can see like the model isn't 100% accurate. We still had to screen 50 to 100 antibodies that were coming out, but we were able from those to find a candidate that was differentiated and that our TPP profile we were shooting for. And we do believe that this is going to be a differentiated asset and profile.

You told us about selection of the targets and the programs that you have. What are the key milestones that are ahead of us in terms of those compounds?

Yes. So for TL1A, we have NHP data that's going to be coming out here shortly, I would say, end of summer time frame. And then we will enter the clinic first half of '25, and then we'll have a Phase I interim readout the second half of '25. So those are kind of the key milestones for the TL1A asset, ABS-101. And then ABS-201 will have a drug candidate by the end of this year. This is for a derm target that we believe is very much underappreciated similar to TL1A. And then we have ABS-301, which is our novel I/O target, which we'll have in vivo validation on that end of this year. And so it definitely is a catalyst-rich year for our pipeline. And then we additionally have 4 new partnerships we're expecting to announce this year as well.

[indiscernible] And I see some of the companies that have products that you're going afterwards have may be inflated or maybe noninflated valuation, but significantly different than where you are. I didn't know the story well, I saw you were presenting, the guys who have worked for me like it. Why doesn't -- why don't people know about your story? It just seems odd to me because the stories that have similar targets are glorified. You probably know the companies I'm talking about.

Yes, like Recursion and -- yes, yes, no, absolutely.

Even the targets for TL1A.

Spyre, yes.

Seeing TL1A.

Yes. No, absolutely. I do think that we are underappreciated. I think that the evolution of the company as well has changed over time, the foundational technology has stayed the same. And I think that there was a recognition of the technology and what the technology was able to deliver with TL1A, and that was, I think, seen in the recent financing that we just did. We released the TL1A preclinical data at JPMorgan and then shortly thereafter, we raised an $86 million follow-on. And we were able to bring in blue chip biotech investors as well as, I would say, more generous growth funds that came in as well. And so I think that, that was kind of a big catalyst for us. And I think we are on the radar of a lot more companies now or a lot more investors. But I would say, yes, we are significantly trading at a discount when compared to some of these other companies. I don't know, Zach, if you have anything else to add?

No, I would completely agree. And as Bartosz mentioned, I used to be on the Board and I thought that the company was extremely undervalued and extremely exciting. And so I was happy to join operations.

Well, I know, you were our first -- so I mean, Zach founded a venture fund. They led our Series A and B and ended up stepping off the Board, stepping down as the general partner of VC he created to come on full time as our CFO and CBO, just recently.

That said another way, first I boarded with my money, and now I boarded with my feet. I think we have a very bright path ahead of us. I think there are some investors who are also still trying to understand AI and what it's capable of. But in this case, we're showing results. We're showing actual TL1A molecule that I think has a profile that's exceptional.

[indiscernible] toxic to some investors, we think that that's just momentum and whatnot. And I just say we actually developed a target before something where there's another company that really just has an idea that has a $3 billion valuation.

Exactly.

So I don't know what the messaging is whether you -- because to me, it's very abstract. Like I said, I sit here and I have guys [indiscernible] looking at Medtronic and United. And then I hear the story, and I can't understand why if you have a TL1A, the investment community is so excited about it in other places, why [indiscernible]?

Yes. I mean look, I think that you are seeing the stock price appreciation, and I think that you are seeing investors get excited. I mean, we're -- we were at $1 late last year. We just continue to execute and now we're 5, 5x that, trading $4.50, $5. And so I think folks are seeing the value creation. And we've really tried to just put our heads down and execute and not really go into the hype and just be like, look, here's our AI model, and here's what it's producing and it's producing valuable differentiated assets.

[indiscernible] data in TL1A?

Yes, we are going to have that. So we're going to enter the clinic early next year. And then second half of next year, we will have our Phase I interim readout. And so we're kind of like right on neck and neck with the other competitor you were talking about.

The partnership structures, will they vary or they have a certain consistency to them?

So at the drug discovery phase, they're fairly typical, encompass upfront payment, R&D funding. And then based on success of the R&D program, so basically for us designing the antibody against the partner's target. Then there's typically an election fee. And then we'd be eligible for downstream milestones, both development and commercial, and we always have royalties in all of our deals. When we look at our asset deals going forward, we expect those to be much more significant in terms of upfronts, milestones and royalties. And this is one of the reasons why we're really focused on developing our own internal programs. We can be very efficient about it. So the ROI is exceptional, assuming we can develop these to a stage where -- and typically, that's human proof of concept where we can generate significant transaction terms.

And so that's kind of on the drug creation starting at the target phase. But like TL1A, we've actually gotten a lot of inbound interest already on that, and we have a data room that's set up where we have large pharma partners that are in the data room. And so when we have that Phase I interim readout, we'll be able to pull the trigger if an offer comes in that we think is attractive. And so we definitely have already, I would say, have curated a nice buyer list for asset that is actively engaged at the current moment.

So in the data room, I presume [indiscernible] competitive nature to it?

Yes, absolutely. There's more than 1 large pharma in there. Yes. We're definitely [Audio Gap].

The state of in-house drug creation versus having collaborations and having the pharma a couple of targets? What should be the goal each year?

I think the biggest driver of value is going to be our own internal assets. But at the same time, there's a lot of value to be created with large pharma because there's only so many assets that we [Audio Gap] and we can't go into every single therapeutic area [Audio Gap] you to have a diversified portfolio that you wouldn't be able to have otherwise. And so we continue to see that be an important driver, but using kind of some of that -- those upfront payments that we get from large pharma to continue to reinvest in our own internal pipeline, where we see the majority of the value being generated in the mid- to long term.

So that goes to the question of runway and funding and balance sheet. So maybe, Zach, you could talk a little bit about that?

Yes. I mean, as Sean mentioned, in early March, we raised an $86 million follow-on and so that gives us runway into the first half of 2027. So we'll be through our Phase I readout. We'll be able to advance the pipeline programs that we've announced [ based ] on our balance sheet today. I think we're in a good position there. And then to Sean's point, we're also looking at different levers for nondilutive capital infusion. So one of those is drug creation partnerships. We expect to do at least 4 of those this year. And the other, which is a little more probably looking at next year and the next couple of years is transacting around some of our asset pipeline. And that's where we think we can generate significant nondilutive capital inflows that can then further fund more portfolio development.

In the past, you pulled trigger on M&A to build your capabilities. How should we be thinking about your capabilities going forward? And how it impacts your M&A philosophy?

Yes, absolutely. We're always looking at M&A opportunity. I think it's always a question of do we build or buy. And I think the reverse immunology platform that we brought in for target discovery [Audio Gap] was important because that allowed us to go not only in the drug design, but also allowed us to go into AI for [Audio Gap]. And then obviously, we brought in Denovium, which brought in the AI capabilities. I would say [Audio Gap] technology is pretty sound, and we don't see any current expansion, but we're always on the lookout for new technologies to add. And if we are going to add new technologies, it's likely going to be on new wet lab technologies that gets us new data, new models that we're working on.

[Audio Gap] here for our group in the room or for investors.

I mean at the end of the day, I think what we've been able to show is that we have a diversified portfolio that we're building that's being driven by our AI platform. And this AI platform has been validated with our preclinical assets like TL1A, but also validated from large pharma partnerships like AstraZeneca, Merck, Almirall, and really, what we want to be when we grow up is really the next Regeneron. When you look at Regeneron, they were the ones that brought in the humanized mouse to produce human antibodies. We see this as the next generation of that and being able to unlock new novel biology, create differentiated assets that are first-in-class and best-in-class. And we're really just getting started. I think we have some really exciting catalysts coming up, both on the partnership front but also on the asset front ourselves. And I think that we're in a really exciting place to go and execute over the next couple of years and have the capital to go do that. And so we're excited about the catalyst-rich year that we have coming up.

Great. Any other questions from the audience? Great. Well, thank you both for coming.

Yes. Thanks so much, Bartosz.

Thanks, Bartosz.