Why don't we get started, everyone? Good afternoon. I'm Steven Mah from the Tools and Diagnostics team, introducing our next company in the AI-driven drug discovery, antibody space. It's my pleasure to welcome Sean McClain, founder and CEO of Absci, and Zach Jonasson, Chief Financial Officer and Chief Business Officer. So, let's keep it as interactive as possible. Anybody has any questions, just either feel free to just ask in real time, or if you want, you can shoot me an email at steven.mah@cowen.com. So, yeah, so let's just kick it off. Thanks for being here.
Yeah, absolutely. Thanks for having us, Steve.
Yeah. So, so Sean, maybe just give a quick introduction to, to Absci, for those new to the story.
Yeah, absolutely. Absci is a generative AI drug creation company that is focused on using generative AI to create differentiated biologics. And really, what we're looking to do with generative AI is really break the economics of traditional biotech, where you're investing 80% of your balance sheet into one asset. What if you could actually use generative AI to you know, dramatically decrease the time, as well as the overall cost, so you can create more shots on goal? And ultimately, that is you know, what we've done here at Absci. And maybe Zach can talk a little bit about the recent financing that we've had and some of the assets that we're developing.
Yeah, I'm very nice to be here, and I'm happy to comment that last week, we closed a registered direct financing, raising $86.4 million. That financing was heavily oversubscribed and really gives us fuel to drive our internal programs forward over the next several years. So that'll provide runway into the first half of 2027, and should allow us to reach phase I readout for ABS-101, our TL1A asset, as well as bring ABS-201 and ABS-301, our other pipeline assets, fully to IND. In addition to that, we'll be developing other internal programs, at least to candidate stage.
Yeah, and I think, with our own internal pipeline, what it's really shown is that, you know, generative AI can actually create, you know, differentiated assets in a very short amount of time. It took us, 14 months to generate a drug candidate for TL1A, and we'll have it in the clinic within 24 months. Now, traditionally, that takes 5.5 years. So going from 5.5 years down to 24 months is a huge time savings. And then, you know, additionally, it cost us, roughly $15 million to get into the clinic, to do our phase I, you know, roughly $5 million.
So to have a phase I readout, with, you know, $20 million of invested capital in roughly three years, we see this as a big paradigm shift in how you're developing assets. And not only are you able to do it faster and cheaper, you're able to actually create a differentiated profile as well, and that's what we demonstrated with this TL1A asset. You know, not only did we, you know, develop it cheaper and faster, we're actually able to show that we could create a best-in-class asset there, which we presented this data at J.P. Morgan, and I think this is what really got investors as well as partners excited and really what led to the overall financing.
But this is like a real demonstration of how, you know, you can go from this hype of generative AI and showing that it is an actual reality. It is here, and we're really excited to be using this to build out our own internal pipeline, as well as drive partnerships with leading large pharma like AstraZeneca, Almirall, and Merck. I think these are huge validations of the platform as well. But, you know, it's been 18 years, or sorry, 18-24 months of really strong execution, and I think we've now, you know, delivered on some of the realities of what generative AI can ultimately bring to biotech and drug discovery.
Yeah. Yeah, we'll get into AstraZeneca and Almirall in a second, but, you know, maybe let's talk about your technology platform, you know, the ACE Assay platform and how that, you know, how it was really enabling you to kind of make that leap from generative AI versus, you know, helping-
Yeah
... just to, you know, maybe optimize kind of, you know, lead development. You know, what was able to make, 'cause, you know, I, I think you're one of the maybe only publicly traded companies that actually in the generative AI space in terms of drug development or, or one of the few.
Yeah. No, absolutely. You know, we have a firm belief at Absci that those that are gonna win in the generative AI drug creation or drug discovery space are those that can generate differentiated data sets and integrate that with the AI. And what we've been able to you know show you know over the past few years is this integration, being able to have a scalable wet lab technology that can you know generate you know millions to billions of protein-protein interaction data points to train our AI models. But then we can also use that same technology to then go into the wet lab and validate that.
We can actually validate over 3 million unique AI-generated AI designs in a given week, and that whole cycle time is about 6-week time period. And that ultimately allows us to almost act like a tech company within biotech and has ultimately led to the breakthroughs that we've had on the de novo foundation model, which led to the TL1A asset as well as some of the other assets in our pipeline. But again, it's being able to develop these differentiated data sets. And look, we are doing innovation on the AI side, but the AI models ultimately are gonna get commoditized, and those that have differentiated data sets are ultimately gonna be those that ultimately win in the space.
And, Zach, if you wanna talk about actually how we generate that data and what makes us unique?
Yeah, I mean, fundamentally, it's a data problem. We're not computer limited. So at Absci, we've developed-
Well, we are. I mean, computer. We do need computer.
We need compute, but that's not-
We are not-
... that's not the bottleneck. The bottleneck is data to produce assets that have value. At Absci we use, as Steve mentioned, we have a proprietary assay system called the ACE Assay, where we do that at scale, and that's what Sean was referring to. We also combine that with other lower throughput data as well as public data. What we're doing at Absci is really building a data moat, because we think that's what's really differentiating, and it's a sort of, if you look at it, as our unfair advantage. It's building, and continue to build that data, those datasets so that we continually improve the models that we're working on in the AI team.
And so it's a case where we have a leadership position that we're looking to extend and continue to extend by building more and more data. And the other comment I'd make about the data component is, it's not just the scale of data, it's having the right data. And so with the ACE Assay, we're really able to dial in and get very very detailed data around the protein-protein interaction, so the antigen interaction with the antibodies that we're creating, and that's really the workhorse for the AI creation that we're implementing. We're using that data at scale, and then we refine that for developability with other datasets that are lower throughput.
Right. And so that, you know, that sort of de novo platform is kind of what drove the Almirall collaborations and AstraZeneca, if I'm correct. As, you know, I know, you know, you're kind of limited on how much you can say about the TL1A asset, but, you know, what gives you the confidence that the, you know, the generative AI platform, you know, works? And I know you have a, you had a pre-print publication-
Yeah
... out there. Maybe just, you know, spend a few minutes on that, if you could.
Yeah, we've shown now that we've created this foundation antibody model that can generalize to, to, you know, various different targets. We've shown in a manuscript that we could, you know, generate, you know, antibodies towards HER2. Now, we've shown, you know, TL1A. We have other examples, and one of the things that is very unique about this is the epitope specificity you get out of the model. No other, you know, technology exists out there right now, where you can have epitope specificity, whether you're, you know, using immunization or phage display or yeast display. And this is really what got AstraZeneca and Almirall and Merck really excited about the platform, was you're able to now go after these undruggable targets like GPCRs or ion channels that don't...
You know, these targets don't stick out like a sore thumb. You know, they, they, you know, they almost stick out like a, a thumbnail on the surface of, of the target, which actually makes it difficult for the immune system to generate antibodies towards those. But if I have an AI model that can, you know, be epitope-specific, it doesn't matter how much surface exposure there is, there just needs to be some surface exposure there, and we can then generate antibodies to these targets and unlock new novel biology. And so that's really what, you know, AstraZeneca, Almirall are really excited about, is using this technology, using this generative AI platform, to unlock new novel biology.
You know, we're gonna continue to see that evolve, not only for our partnerships, but also for our own internal pipeline as well.
Okay. You know, appreciate the color. You know, the pharma industry, maybe let's talk about, you know, that a little bit more. You know, I see, you know, them, you know, moving into the AI space. You know, what's your sense for, you know, AI's... or, sorry, pharma's appetite for, for AI going forward?
Yeah, I, I think that pharma's appetite is, is extremely high. What you're seeing is pharma is building out their own internal capabilities, and then they're figuring out where do they have gaps, what, what problems and data can, you know, they solve themselves and, and use, you know, AI to solve these kind of big problems? And then where are areas where they're data limited and need to partner to kind of solve some of the, you know, some of these problems? And I think that's what you're seeing is, is pharma, you know, looking to partner with, with Absci, where, let's say AstraZeneca, you know, needs a solution for antibody drug discovery, but they have really great clinical data for, you know, using AI for, let's say, you know, getting the right patient enrollment in a clinical trial.
So I think you're gonna see this, you know, both this, you know, build and kind of buy mentality within pharma. But pharma, I think, has seen the benefit of it. I think pharma sees that AI is going to stay, and I think some of the proof points that even Absci has shown with the TL1A asset is even getting pharma even more excited. And I think one of the things that we see as really important is that we aren't competing with pharma with the assets we develop. We never plan on taking any of our assets past phase II clinical trials, so we're never gonna compete with pharma on the commercialization side. We're actually leveraging pharma with the skill sets that they're really great at. They're great at late-stage, you know, clinical development.
They're great at marketing and commercialization, and we're, you know, helping them generate, you know, assets and taking them to a certain stage, and then partnering them at that point in time. So we see it at the, our business model and how we interact with pharma as very synergistic within the ecosystem that is already created.
Okay. No, no, appreciate that. And yeah, maybe that's a good, good segue into my next question, you know? Yeah, we totally understand your R&D partnership business model, but, you know, the pivot now to your kind of internal programs, totally owned programs, you know, how do you guys, like, weigh the puts and takes of that? You know, there's obviously gonna be increased cash burn-
Yeah
... and, you know, you're not gonna be receiving upfront payments or kind of, you know, operational R&D milestones as you would if it was a partnered program, you know?
Yeah.
Maybe, maybe your thoughts around that.
... Yeah, look, let's just use this last JPMorgan as, like, I think, a great case study for, you know, how a pharma's starting to look at this. Yeah, pharmas, you know, patents are expiring. You know, a lot of the late-stage clinical programs are being picked over, and they're reaching early and earlier into the clinical pipeline to acquire assets. I think the Harpoon/Merck acquisition is a perfect example of that. That was a phase I oncology asset that ended up getting picked up for $600 million.
And if you can now generate, you know, a candidate that can go into the clinic or even into a phase I, and you're able to do that for $20 million and, you know, call it 2-2.5 years, and sell that for, you know, $600 million or more, the... You know, you're starting to, you know, get to a point where you can start to have a new business model. So instead of investing that $100 million to get there, you're now investing $20 million to get there, and you have a differentiated asset.
And so you're able, with kind of the a similar balance sheet as a typical biotech company, invest in, you know, not one asset, but five assets, taking them to, you know, different value inflection points. But again, with our own assets, we never plan on taking them, past phase II clinical trials. We ultimately will partner these out, but I think it does make sense to take them later on, especially with how cost-effective we are of getting them to these different value inflection points. I don't know if, Zach, you have anything else to add.
Yeah. I mean, Steve, I would say if we just look at the ROI, you know, Sean just cited the demonstration of what we did with TL1A. We believe the platform is getting more and more efficient as we go through time. We're adding more data to it, we're refining the models. So if we can generate assets at that speed, at that quality, and that cost effectively, there's a really great ROI there. And then it becomes an exercise in making sure we select the right targets, and we have a wonderful team that works on doing that. But I would say hand in hand with that, we do expect to continue to do partnerships with pharma on the drug creation phase.
In addition to bringing non-dilutive capital in, there are great synergies there with respect to what the pharma partner brings and their biology expertise, right? They may bring domain experience in dermatology, like Almirall, and we can leverage that and have a synergistic partnership. So we're looking to do both, and in that way, build a diversified portfolio.
Okay. All right, great. So yeah, let me just pivot over to kind of your lead, wholly-owned asset, TL1A. You know, can you, you know, give us kind of, like, a high-level overview of what you guys reported back in January? Zach, you want to-
Sure. I love talking about it, so I'm happy to. Yeah, I think we're really excited about this, and I think investors had a strong response. We've had a very good response from pharma as well. But we've generated an asset in this space that we believe is highly differentiated. To begin with, we've designed an antibody against an epitope, so that's one unique feature about what our AI capability brings. We can design towards a specific epitope or epitope region. So we've selected an epitope in TL1A that we believe will have a lot lower immunogenicity, so a low ADA rate, and that would be as compared to what the Roivant molecule and the epitope there is involved in. We believe that that ADA experience that they've had in the clinic is driven by the actual epitope interaction with the antibody.
So we've been very careful to select the epitope. Secondly, we've designed in a longer half-life, so we've done Fc mutation. Our expectation and our goal is to have a once-a-quarter dosing profile, which we think will be very differentiating for this market. And then finally, I would mention two other factors. One is we've designed a molecule that has higher potency than anything else that's ahead of us in the clinic. And then finally, it has a very good developability profile, so we expect to be able to formulate this at high concentration in a low-volume dose that'll be very convenient for patients.
Right, and you, you're making these comparisons based on their kind of preclinical data that's in the public domain?
Yeah, and we've done head-to-heads, at least with the Merck and the Roivant molecule, because we know those sequences. I think we've just sort of now been able to identify what the Teva sequence is, so we've started doing some head-to-heads there as well. But all of our potency and comparative data has been done against those sort of innovator molecules.
Okay. And you got the sequences, like, from the patent?
Yes.
Okay. All right, no, just curious how you got that. Okay.
Yeah, it was-
And then do you then express that on your ACE Assay, and then Did your comparisons that way?
We had those ones made in CHO cells, so they had the glycosylation. Oh, okay. Yeah. But yeah, we had all of those made, and then we did a head-to-head comparison with the cell-based assays. And then we also did the mouse PK studies as well. Okay, comparing the half-life. Okay, got it. Okay. And then, you know, what should we look for in terms of, you know, TL1A data kind of readouts over? I know you're doing IND-enabling studies now, but, you know, are there gonna be any kind of readouts on your? Yeah, preclinical data or PK studies?
Yeah, so we'll have PK studies on NHP that will come out this year, along with the clinical trial plan that we have- Okay ... for TL1A. And then the plan is to wrap up the IND-enabling studies end of this year, have an IND first of 2025, be in the clinic shortly thereafter, and then in the second half of 2025, we will have a phase I interim readout that we'll be sharing.... and then the clinical trial should be wrapped up in Q1 2026, or sorry, 2026.
Okay, all right. Got it. All right, that's very helpful. So maybe, just maybe, I mean, pivoting over to, you know, kind of generative AI, and I mean, sort of, kind of hopping over here. But you know, do you expect to see, like, more companies getting into generative AI space? Or is it really kind of limited by, you know, the ability to generate, you know, really high-quality functional data? Or do you expect there to be kind of like the, that kind of next big thing, and everybody's gonna be moving into generative AI?
Yeah, I think it all comes back to data and compute. Do you have access to differentiated, you know, data sources? You know, whether you're buying those data sources, you have technology that can generate those, and then do you have the access to the compute that's necessary? And you really need both of those in order to be successful. And then ultimately, how are you then validating the model? So you can get the, you know, these differentiated data sets, train the models, but then you have to then go and validate them.
That's why I think it's really critical to have this iterative loop where you have scalable technology in the wet lab that can generate the data for training, but then once those models are trained, you're then going into the wet lab and validating them. And, you know, again, we can validate over 3 million unique AI-generated antibody designs in a given week and see how accurate the models are. You know, should we be doing different, you know, hyperparameter tuning? Do we have the right model architectures? Do we have the right, data? So that ability to generate the data, but then also to validate the models, and being able to do that in a short time period, I think is absolutely critical.
You know, we can do it in a 6-week time period, but ultimately, to build these, you know, accurate foundation models that are highly generalizable, you absolutely need that iterative feedback loop. And I think those that are reliant only on publicly available data and, you know, and thinking that an AI model alone will create differentiation, you know, ultimately, we do not believe that that's the winning strategy. I think that you do need that differentiated-
Yeah
... data. And then once you have that differentiated data, then it's, you know, I think compute really comes into play. How can you scale that, you know, these models to ultimately increase the overall accuracy as well?
Mm-hmm. Okay, all right. That's helpful. And yeah, maybe on topic of compute, you know, what's your thought on... I know, Zach, you said, like, you know, the data is actually the limiting—being able to generate high-quality data is sort of limiting factor. But, you know, what do you think about, you know, some other companies in your space, which are, you know, kind of building their own supercomputers versus using the cloud? You know, what are kind of the pros and cons, and, you know, how do you guys approach the decision?
Well, just to qualify my comment before, I think the data has to come first, and then the compute. Like, with biotech, where we are, we need more data, and then we can—the compute can scale with that.
Right.
But as a first-order problem, we need to generate the data, and that's really what's missing in the industry, and that's one of our big differentiations.
Okay. And you guys, right now, do you use your compute in the cloud right now, or is it on premises, or?
Yeah. So, so we have our, our own cluster, which we are, looking to, dramatically scale up to build the, the next, version of the foundation model. And so we are looking to, scale that. I think there's different ways to scale it. You can either go out and buy, you know, GPUs, or, you know, you can use, you know, cloud, cloud services, I think... But, but ultimately, you know, kind of where, where we're at, like, we're looking to kind of take it to the next phase. We've, we've scaled data, and now we want to, you know, continue to, to scale the, the, the compute, more than what we've already invested.
Okay. All right, that's helpful. And then, you know, sticking on the topics of, you know, other peers in the space, you know, you guys have chosen to, you know, take an antibody-based approach. You know, others in the, in the space have decided to do a small molecule approach. You know, kind of what are your thoughts on the, you know, the, the different drug modalities and, you know, and, and being able to apply AI algorithms, to small molecules versus biologics?
Yeah, I mean, I, I would say that in the biologic framework, it's, it's more compute and data-intensive to design the actual molecule, and that's really what we're solving for, and at the same time, we're leveraging a lot of the inherent specificity of an antibody, for example, right? Like, I think there's some recent studies that look at success rates or probability of success for antibodies versus small molecules, and they range from 2-4X better probability of success for an antibody, and that's because of the specificity that's endemic to that kind of a, a molecule, and it's, it's, you know, basically less toxicity. So we're leveraging a lot of that, but it's very compute and data-intensive to actually do the design of the molecule. I think on the small molecule side, the problem is a little different.
A lot of it is really, how do you model out the off-binding effects or what the promiscuity of that structure is gonna be? And there, I think it's challenging to get the right data sets in place. So in some ways, you know, we're solving different problems. There's a Venn diagram that overlaps, but in some ways, there's a different problem set that the small molecule players like Recursion need to solve versus what we're solving.
I see. How much does, like... I mean, using, you know, obviously, you know, evolution has given us, you know, kind of a, kind of a roadmap there, and I, I know a lot of the, a lot of the models use, like, a, you know, evolution-based transformers, where, you know, small molecules don't have an advantage of that. How, how much is that, an advantage, being able to start out with sort of, you know, nature kind of already having a, kind of roadmap for you guys?
Yeah... I mean, I think it helps out a ton on the immunogenicity. I mean, I think that that's a big, you know, concern in, you know, with small molecules, is, like, the off-target effects that you get.
Right.
You know, antibodies are very specific to the target of interest, and there's not as much off-target effects that you see. And then also, too, an antibody is natural. Now, you do have to take into account when you're designing the CDRs, are they as human-like as possible to really ensure lower immunogenicity? But then there's also, you know, items like you just can't control of, like, you know, how does it bind to a complex? And, you know, you have different B-cell-mediated responses that are, you know, definitely hard to ultimately predict. And I think these are challenges that, you know, we're ultimately gonna solve with AI, but we're not there yet.
But I think that when you compare a large molecule to a small molecule, I think that there are a lot more advantages just from the immunogenicity standpoint and off-targeting effect. And then also, too, I think IRA is definitely putting a big spotlight on R&D investment within small molecules.
Mm-hmm.
I think you're seeing pharma change their R&D budget to focus more on, you know, other modalities, you know, antibodies, cell and gene therapy, than you know, continuing to make the investment in small molecules.
Okay. All right, that's really helpful. And Zach, you know, we only have about five minutes left. Let's talk about on the business development side, program partnership funnel. You know, can you tell us... you know, give us kind of from a high-level perspective on, you know, the level of inbound you're getting from, you know, potential partners. And, you know, kinda given kind of the confusion and, you know, buzz around, you know, AI and generative AI, you know, is all that noise helping or hurting the conversation?
I mean, in some ways, especially last year, it slows things down a little.
Yeah.
Because there is a lot of noise. You know, I have a friend who always says, "There's a lot of companies where it's AI sprinkled on top.
Yeah.
That's, that's not us. We're AI to the core, and we've built it on top of a data engine. But what we are seeing is pharma... You know, we had—you had this question earlier about what pharma's thinking about AI. A couple of years ago, I would've said a lot of pharma executives were sort of wondering if, you know, if, "Is AI actually gonna impact and transform drug discovery?" Now, it's really a question of when.
Gotcha.
So they're investing in their own internal teams. I think what we're finding is, they're now at a place where they can do diligence and understand what's different, and that's the case with AZ. Our partnership with AZ was a year plus of diligence, where they looked at every company in the field and decided to work with Absci. So we're starting to see pharma have the capability to really do the fundamental diligence. So, back to your question, how does our pipeline look? It looks better than it's ever looked before. I think the industry paid a lot of attention to the deals we announced last year-
Yep
... as well as the TL1A data that we've released. And so we haven't. We'll put out some guidance later this month on the number of partnerships we expect to do this year. But I would say the pipeline looks strong, and we're looking to do more kind of thematic-type partnerships with pharma-
Okay
... gonna go forward.
Got it.
Again, I think pharma, what they're seeing is that we—you know, we're actually showing how generative AI can make a difference, how you can create a differentiated asset in a timeframe that hasn't been seen before, at a cost that hasn't been seen before. And that's, you know, really driving the interest, and it's what drove the AZ deal. But then they're also interested in the pipeline assets that we're developing. You know, our TL1A asset, our, you know, first-in-class IO target, our, you know, best-in-class derm target. And so we have—I think the strategy we now have is multi-pronged, where you can... You know, you're partnering both on the asset side, but also on the platform, you know, target-based side as well.
And so I think it creates more opportunities for, you know, bigger, and more in-depth partners, where partnerships, where you can invest. You know, you partner an asset off, but then they're diving deeper, maybe in a therapeutic area. And, you know, you're co-developing, you know, 5 targets together in addition to the asset. So I think it just creates more partnership opportunities. And we've seen just an uptick in inbound interest on the pipeline assets, but also on the platform itself.
Okay, gotcha. All right, so it sounds like it wasn't some killer app, where people were like, "Okay, this is real." They actually got in the weeds and kicked the tires of, of Absci before deciding. Is that, that kind of a fair way to look at it?
Yeah, you could use the analogy of a root canal. I'm kidding.
Okay.
But they did a fair amount of diligence-
Yeah.
Yeah
... for sure.
All right. All right, cool. Well, in the last minute or so, you know, any other exciting catalysts? I know we talked about, you know, 202,
Yeah
... you know, or sorry, 201, 301. Anything else looking forward to-
Yeah
... this year?
Yeah, so, yeah, just to recap this year, we'll have NHP data on TL1A. We'll have IND studies completed. We'll, for ABS-201, which is a best-in-class derm target, have a drug candidate on that, and then we'll be disclosing the preclinical data shortly thereafter. And then, ABS-301, a first-in-class IO target, we'll have in vivo validation on that this year, and then a drug candidate shortly thereafter. And so those... And then additionally, we do see more AZ-like partnerships being announced this coming year as well.
Okay.
So I think those are some of the, you know, catalysts that we see coming up in 2024.
All right. Well, fantastic. Yeah, looking forward to the earnings call then.