Morning, everyone. My name is Gil Blum, and I'm a senior analyst here at Needham & Company. Thank you all for joining me on this day of the Needham Growth Conference. It is my pleasure to have with me today Sean McClain, the CEO of Absci, and he will tell us all about Absci's AI platform in biologics. And Sean, with that, please go ahead.
Awesome. Thank you so much, Gil. I'm Sean McClain, the founder and CEO of Absci. We're a data-first generative AI drug creation company. And what we're doing here at Absci is really revolutionizing the way we design biologics, being able to design in attributes that allow you to, you know, create first-in-class, best-in-class assets. And we've had a ton of success and momentum over the last year, and, you know, some of the recent highlights are highlighted here with... Apologies. One sec. With AstraZeneca, we closed over a $250 million deal within oncology. With Almirall, we closed over a $650 million deal focused on dermatology. We closed or met our guidance for 10 new active programs, and, sorry. Sorry, my screen went blank here. Sorry, I'm just having a little technical difficulty. Perfect.
Sorry about that. Yeah, so we've had a ton of momentum over the last couple of months. Recent successes, we've closed over $1 billion in overall pharma deals, two really exciting partnerships, one with AstraZeneca, the other with Almirall. AZ was focused on oncology, Almirall focused on dermatology, and this was really centered on our de novo AI platform. Late last year or earlier in the year, we came out with this model that showed that you could actually design an antibody from scratch.
This year, we took it a step further and showed that we could actually apply that to our own internal pipeline, TL1A, to show that we could create a best-in-class asset, and that's what was really driving these new partnerships, not only with AZ, but also with Almirall. We ended up meeting our overall guidance for the year, meeting two to 10 new active programs for the year. Then, as I said at R&D Day late last year, we rolled out our own internal pipeline. Our lead assets focused on TL1A, using our de novo AI platform to create a best-in-class asset for antibodies.
This is really the first time anybody has actually shown that you could create a best-in-class asset or differentiated asset using generative AI, and I'll be going over that data here later in the slides. Everyone talks about AI drug discovery. Now, what is separating us from a lot of the others out there is this integration between the wet lab and the dry lab, or the wet lab and the AI. It's data to train, AI to create, and wet lab to validate. If you look at GPT-4, it's trained on the whole internet, and if you look at the biological data that's available, that's publicly available, it's a drop in the ocean compared to that. So you need to generate your own proprietary data in order to train these models to ultimately get the accuracy and the output that you want.
That's what we've done here at Absci. We've created this platform that allows us to generate data for training, and then also using that proprietary wet lab technology to validate the models to see how accurate they are. This is ultimately what has led to the success of our de novo AI model we've created. First, data to train. We have an exciting SoluPro cell line that has allowed us to scale the production of antibodies. Normally, with producing antibodies in CHO cells or transient cells, you can only produce maybe 1,000 in a given week. In our E. coli cell line, we're able to scale that up to producing millions of antibody drug candidates at a time to screen.
And then we screen that through our proprietary ACE screening assay, where we're able to interrogate every single E. coli that's producing a different antibody and look at how it's binding to that particular target of interest. You know, what epitope is it binding to, and what affinity? Essentially, the functionality of this particular drug of interest, and this is the functionality data that ultimately gets fed into our proprietary AI models. We take these datasets, plus publicly available data, and again, that's what's used for our training of our de novo AI models, as well as being used in our lead optimization models.... Earlier last year, as I mentioned, we came out with a really exciting new breakthrough AI model in the space of antibody design.
This is a foundation model for being able to create new antibodies from scratch. And how the model works is you take a structure of the target, whether it's a crystal structure or from AlphaFold, you feed that into the model, and you specify what epitope you want to bind to and what framework sequence you want. And the model's then able to design the CDR regions, essentially the areas of the antibody that bind to the target of interest. And the model then is able to generate sequences that bind to that particular epitope of interest. And we demonstrated that this is actually possible in a manuscript that came out again earlier last year. Now, the last component, and this is a really critical component, is being able to use the wet lab technology to validate these AI models.
Since we don't have a ground truth, we actually need to experimentally validate whether these antibodies have the drug attributes that we want. And so we use that same E. coli strain, the SoluPro strain, plus the A screening assay to validate, but we also have other functionality and developability screenings that we do as well. And this whole cycle time is roughly six weeks, going from the data to train AI to create and wet lab to validate. So we're able to rapidly iterate on our model designs and architectures. And again, this is really what's led to some of the success, is just how rapidly we can move. And additionally, we're able to validate over three million unique AI-generated designs in a given week.
This has allowed us, again, to rapidly move, and we've applied this whole process and the de novo AI model to TL1A to create a best-in-class asset, which I'll be talking about on the following slides. Now, we're not just applying the AI to de novo design, we're applying it across all of drug discovery, from target discovery to de novo design to AI lead optimization. And so what is this enabling for Absci and our pharma partners? First, it's allowing us to create new novel biology, being able to go after these undruggable targets, such as GPCRs or ion channels. We're able to also do multiparametric screening. Let's say, pH, you want to have pH dependency. You want to have the developability and manufacturability with the right potency.
All of this is going to allow you to increase probability of success by being able to really go from this paradigm of searching for a needle in the haystack to creating the needle with the right attributes the first go-around, eliminating this trial-and-error process. Additionally, we're able to reduce the time it takes to get into the clinic. Normally, it takes five and a half years to get into the clinic. We're able to do it in roughly 18-24 months, and that's to get an IND. We're able to do this in, you know, roughly $14 million-$16 million, which is significantly lower cost than what has traditionally been seen.
And then last but not least, not only are we able to use the de novo AI to access new biology, we're also able to use it to access new intellectual property. By being able to screen a much larger search space, we're able to get much larger breadth of claims. And so since, again, we're able to search that large search space and then go into the wet lab and validate over three million unique designs that have that sequence diversity, that allows you to have much greater breadth in your claims and actually enable those claims that'll enable you to have broader IP. So you're going to be able to have enhanced IP production due to our Integrated Drug Creation platform.
There's been a ton of momentum in our partnerships lately. We've closed, you know, roughly $900 million of overall deal value with AstraZeneca and Almirall. And, you know, if we take a look on the right-hand side, this is a nice illustrative graph of how these partnerships are structured. So we get upfront payments in addition to these pharma partners paying for the R&D work. And then from the you know, once they license the molecule, we get a exercise fee, and then additionally, we get clinical milestones and commercial royalties. And again, this is an illustrative purpose shown below here. And so we're getting paid along the way, and ultimately, we get to share in the overall success if the drug is approved with sharing in royalties.
Not only are we partnering with leading pharma companies like AZ, Almirall, Merck, but we also have really important data and compute collaborations, as seen on the left-hand side. In 2022, we announced an exciting partnership with NVIDIA to help us scale our compute for our AI de novo model. We also have data partnerships with leading institutes getting patient samples for our reverse immunology platform to discover new novel targets. And actually, ABS-301 is a novel target that did come from our reverse immunology platform. And so with that, let's dive into our proprietary pipeline that we have. So we have three assets in our own pipeline. Two are best-in-class, one is a first-in-class. The first is TL1A. The indication we're going after is IBD.
The second is a dermatology indication, an undisclosed target, ABS-201. And then last is the novel IO target that we discovered from our reverse immunology platform, ABS-301. And I'm going to be highlighting today some exciting preclinical data that we got and announced at the J.P. Morgan Healthcare Conference last week. So a lot of you know that the IBD market is a very large market and an underserved market. I won't spend too much time in this. There's been also some really exciting deals within the TL1A space. So jumping in. What we wanted to do was take TL1A and use our AI de novo platform to design a potential best-in-class asset towards TL1A.
Now, this is the first time that we've applied our de novo model to an actual asset, and I'm excited to share that we have been able to demonstrate that we can create a superior preclinical profile, and we're really excited to be taking this into the clinic. So first off, we're able to design a molecule that, compared to the competitor molecules, had higher potency. We're able to demonstrate we could have extended half-life, allowing us to potentially have longer dosing intervals. We plan to have sub-Q dosing, and then also have favorable developability. And then additionally, we're able to create some new novel biology as well, which I'll get into, which is in regards to immunogenicity.
And then last but not least, we're able to create differentiated biology. We're able to get around existing IP claims of competitor molecules and carve out our own IP landscape due to our de novo AI model. So let's dive into the data. On the left-hand side here, we looked at our three advanced leads that we have, ABS-101 A, B, and C, and compared those to the competitor molecules from both Roivant and Merck. And as you can see, compared to the Merck molecule, we do have increased affinity, and we have equivalent affinity compared to the Roivant molecule. Now, how does this translate to potency? We can see that here on the right-hand slide. We did an apoptosis inhibition assay in TF-1 cells.
You can see that we were able to achieve superior potency for when compared to both the Roivant molecule and the Merck molecule. So we were again able to achieve that increased potency all driven from our de novo AI platform. Now, one of the profiles that we did want to change for a potential best-in-class was being able to extend the half-life for longer dosing intervals. We did some in vitro Fc recycling assays shown on the left-hand side here, and you can see that we do have increased recycling compared to the competitor molecules.
And then on the right-hand side here is some preliminary in vivo PK data that shows that we do indeed have a longer half-life, in our PK studies, when compared to the Roivant molecule. Again, these are preliminary results. The rest of the PK study will be coming in here shortly, at the end of the month. And then from there, we'll select the candidate to be moving forward into IND-enabling studies. But what this shows is that, it does look indeed like we will be able to extend half-life, and improving the dosing intervals. All right, now getting on to how we used the AI to create some differentiated biology that we believe will help with the immunogenicity.
So one of the things that we found was really interesting was the Roivant molecule, RVT-3101, highlighted in the blue on the left-hand side, binds to a different epitope than Merck. And in the clinical studies that are out there, the Roivant molecule has a higher ADA response than the Merck molecule. And since this is a chronic disease, you know, this could be a concern in late-stage clinical development. And so what we wanted to do was use the AI model to design a antibody that bound to the Merck epitope, but had the potency of the Roivant molecule. Because we believe that the immunogenicity is actually B-cell mediated, and so it's caused from the actual structure.
If we could hit the Merck epitope, but again, have the potency of the Roivant molecule, we could have a superior biological profile there, having the high potency as well as the low immunogenicity. That is indeed what we did. You can see where the three antibodies we created bind relative to both the Merck and the Roivant molecule. Then additionally, this allows for IP differentiation, as well. So using the de novo AI model to create differentiated biology as well as differentiated IP. This is how, I mean, this really demonstrates how you can use AI t o really engineer the antibodies in a way that you could have never been able before .
I mean, you know, immunizing a mouse, you have no control over the epitope or where it binds. But with this AI model, we know what biology we're trying to achieve, and we can then engineer that into the molecule, which is really exciting, and I think the first time that this has really been demonstrated. Now, what does this profile look like compared to the competitor molecules? Well, we see that we will have a better immunogenicity profile, higher bioavailability due to the potency. Also have sub-Q injections, and then going from once monthly dosing to once quarterly. And we do believe that this will be a superior best-in-class profile. Now, where are we at on the timelines?
So we will, as I just mentioned, we'll be selecting our development candidate at the end of this month. We'll be initiating our IND-enabling studies in February, and we'll have the IND submitted in early 2025, and then starting phase I trials shortly thereafter. And what I will note is that it took us only 14 months to get to a drug candidate, and all-in to get to an IND, it'll be roughly 24 months. So we're gonna go from five and a half years normally, down to 24 months. And not only is this speed, I think, an incredible thing to highlight, but it's speed plus that ability to create that differentiated profile.
And I think that this is really what is going to change how we do antibody drug discovery. Now, it's not only just the AI model that we have that's really propelling us forward, it's also the human intellect as well that's going into this. We have an extraordinary team. I like to say that we have a bilingual team, folks that understand the AI, but also understand the biology. We have incredible drug hunters that are here, incredible synthetic biologists, AI scientists coming from leading institutes, as you can see on the right-hand side here.
And not only that, we have a state-of-the-art, 77,000 sq ft campus, that's in Vancouver, Washington, where we do all of our wet lab data for the training, as well as for the validation of the model. And last, I'll highlight the executive team that we have here. Over the last year, we've built out truly an extraordinary executive team and board of directors. We recently had brought on the head of R&D from Bayer as well as Shire, Andreas Busch, to the team. Zach Jonasson has recently joined us as our CFO and CBO. And then additionally, Andreas has built out his team with Amaro and Krishna and Christine, all industry vets.
Then recently, we just brought on Mene Pangalos. He is the Executive Vice President of R&D at AstraZeneca. And so you can see from here, again, we have this mix of folks that really know the AI and the tech, but also are experienced drug developers and drug hunters. You need, again, that bilingual and that kind of having that team mentality in order to accomplish what we're looking to do is combining the AI with the biology to ultimately create better biologics for patients faster. So in summary, through this presentation, I think we've demonstrated that this technology has the potential to create better biologics faster at lower costs.
We're able to advance these exciting AI models because we have the wet lab integration, the data to train AI to create and wet lab to validate. We can do that all in a six-week time period, and we have validated the platform through industry-leading partnerships, most recently with AZ and Almirall. And last but not least, we have rolled out a really exciting internal pipeline, showing how we've been able to create a potential best-in-class TL1A asset from our AI de novo model. I think the first time anyone has really shown how you can use a de novo AI model to create a differentiated asset. And then additionally, we do have some exciting first-in-class assets to really show that we have a diversified portfolio as well.
So if we look at the catalyst coming up for 2024, just like the end of 2023, I think we're gonna have an exciting catalyst-filled year. Both on the partnership side, I think you're gonna see more AZ-like partnerships coming in 2024. In addition to data-rich catalysts, we'll have our IND-enabling studies wrapping up for TL1A, with the IND submitted early 2025. For our first-in-class IO target, we'll have in vivo validation on that, and then with our ABS-201, the derm target, we will have a development candidate on that. And then, as I've mentioned, we'll have the IND submitted for ABS-301 in early 2025, entering the clinic shortly thereafter.
And so with that, I'll hand it back over to you, Gil.
Thank you, Sean, and thank you for that clear presentation. We'll take a minute here to poll for questions from the audience. So in the meanwhile, while we're waiting for additional questions from the audience, basically, the target that you've shown was already a known target, but are there ways that Absci's platform could be used to identify novel biology, meaning new targets?
Yeah. No, absolutely. So I didn't talk about it in this presentation, but we have a reverse immunology platform that's focused on TLS biology, tertiary lymphoid structures. We acquired this company, Totient, that had been working on this platform with Daniele. He was a professor at Oxford, where he was really able to show that these tertiary lymphoid structures in tumor samples have a very different B-cell antibody repertoire than what's in peripheral blood. And if you take these patient samples from those that have had an extraordinary immune response, and you computationally reconstruct the antibodies in those tertiary lymphoid structures and do a proteome panel screen, you can actually find new novel exciting targets to go after, both in immuno-oncology, but also in immunology as well.
And that's where ABS-301 came from. That's an exciting new IO target. And what we're excited about that is the biology on that is stimulating the innate immune system, and we've done in vitro validation on that. We're currently doing in vivo validation, and hopefully, around May, we should have the data on that. And we plan to present it sometime this year, assuming the data readout looks good on that.
Maybe a related, then somewhat more conceptual question. So, you know, functions like ChatGPT and others, they have a generative aspect, which is the ability of the platform to identify basically underlying structures. You know, an example for ChatGPT is it learns languages it wasn't taught. So, you know, it can speak in languages that it wasn't taught just because of the underlying structure. Is there an analogy for this sort of development in biologics in human disease, meaning, can an AI identify, quote-unquote, "an underlying structures," which a normal human can't?
Yeah, absolutely. I mean, you can... The AI is able to, again, go from what used to be, I guess, classical AI, where it's classification, where it's, like, identifying, you know, it's trained on a bunch of pictures, and then it can identify what those pictures are, whether it's like an apple, a banana, or a mango. It can successfully identify those. Now, we're going into this new era of generative AI, where you can use these, you know, large training sets, like the ones that we've created on antibody functionality, and then the AI model can search outside of that data set, and ultimately create something new from scratch.
And I think that's the really exciting era that we're in. But in order to make that successful, again, when you're kind of in this creation or this generation of new novel antibodies, you have to be able to validate those in a very rapid manner, 'cause you don't have a ground truth. Like, with AlphaFold, you had a ground truth. You knew what the structure was, and you were trying to use an AI model to get, you know, as accurate as you could to the known structure. But with these, you know, when you're creating a new drug from scratch, you don't have a ground truth, so you have to actually go into the wet lab and validate it.
And I think that that's really, really important to, to, to highlight again, and that goes back to that AI-wet lab integration. I think those that are ultimately gonna be successful in this space are those that have that, AI-wet lab, integration, and, and you're able to rapidly iterate on that. And we can do that in that six-week time period.
Maybe to translate that, the idea is clear out those hallucinations quickly.
Absolutely. Yep.
Okay. So we also understand the development of, you know, biologics to novel targets is the higher unmet need here, you know, all those undruggable targets, et cetera. But given your partnerships, isn't there some potential here for, you know, best-in-class to known targets? You know, do you get inbounds from pharma regarding optimizing their existing portfolio?
... yeah, absolutely. I mean, we have inbound interest from pharma where they actually have known targets that have known biology, but they've been difficult to drug, you know, like a GPCR or an ion channel. I think, like, partnerships like AZ and Almirall really focus in on that. And so those still will be first in class, but you at least have a known, you know, biology around that, and, you know, the likelihood of success, I think is greater going after those, kind of low-hanging fruit, with, again, with kind of the ion channels and GPCRs. And we do have partners that are coming to us, with those.
But additionally, we do have partners that come to us with, you know, their own novel first-in-class assets, as well, that they would, you know, like us to develop drug candidates towards.
Maybe to clarify kind of where I was going with this, you know, there are known targets out there, like Herceptin, you know, for HER2 and, you know, other, other cancer targets. But most of these companies have a patent term on them.
Yeah.
There's always... pharma always looks for extensions of patent term. One way to do that would be to come with a, you know, fast follow better.
Yeah. So there definitely is, like, the fast follower approach on, like, on a HER2. But, you know, speaking of HER2, I'd argue that we actually have an exciting HER2 antibody that we've created. We actually recently just put out a publication on that, where we actually showed that we could get 3x higher potency compared to trastuzumab. But we don't, you know, that's a very crowded space and, you know, if a partner wants to go into there, I think we can make that determination if we wanna, you know, partner. But in terms of, like, our own portfolio and how we look at it, we don't wanna go into a super crowded space.
If we are gonna be a fast follower, we wanna, you know, we wanna come in, you know, maybe, you know, coming in at, you know, second or third and have a really, really differentiated profile versus coming into, I would argue, a very crowded market like HER2. But again, you can use the AI to generate novel IP in there to extend patent terms and, you know, I think that that is an interesting area we can go into. It's not a big focus of ours right now, but it is one application we can go into.
So maybe spending a minute here on the business model. So overall, we didn't talk a lot about this during your presentation, but generally, the company is focused on, you know, phase II clinical development to create the most value, kind of past human proof of concept, and that makes a lot of sense for cash management. However, do you think that kind of puts a cap on how much upside can grow as a company?
Not, not at all. So what we're seeing is that with the AI, it is enabling this new business model where you can get to a clinical readout a lot faster and at a lot cheaper than a, you know, a traditional biotech could. And what we're doing is now spreading that cash that you would normally invest all in on one asset, you know, to a handful of assets. So you have more shots on goal, and theoretically, these are gonna, you know, be better-designed molecules than you could traditionally get, so you're gonna likely have higher success on those.
And you can take those to, again, you know, clinical proof of concept, or even taking it to, to an IND and, and bringing in that kind of longer-term cash, bringing in, in, you know, sooner and, and being able to then, you know, sell that asset and kind of rinse and repeat. And, you know, we see pharma, you know, reaching earlier and earlier in, in, in the pipeline, to, to get these assets. I mean, I think Merck Harpoon acquisition is, is a perfect example of that. I mean, a phase I asset that, you know, sold for $600 million.
If we could take, you know, ABS-301 into a phase I and sell it for, you know, $600 million and use that cash to reinvest into our own portfolio to do that all over again, I think that that, you know, makes makes a lot of sense. And some of these, we may decide to take, you know, to a phase II and make that determination. It's worth it for us to, you know, go from a $600 million acquisition to, you know, a couple billion-dollar acquisition. And so these are just the you know, it's a case by case, but I will say that we won't act like a traditional biotech, where 80% of our balance sheet is gonna be focused on one asset.
It's gonna be focused on this diversified portfolio that we, you know, ultimately, take to proof of concept. And I will also say, like, you know this, Gil, like, the pharma industry, it's really a team sport. You know, we know what we're good at, pharma knows what they're good at, and if, you know, I can, you know, say with a straight face, we're not set up for late-stage clinical development. We're not set up for the, you know, the manufacturing, the sales. Like, pharma's great at that, and so-
500 patient plus phase III. Yeah.
Exactly. So let them do what they're good at, and we can do what we're good at, and, and, and that's really, I think, what, you know, what we're focused in on. And I think, and we're able now to, you know, take these to value inflection points, at a cost that you couldn't have done prior.
A related strategy question here is, so is there a situation, you know, if the indication or drug is the right fit, let's say? You know, a rare disease, would you consider full clinical development?
So I would say our Chief Innovation Officer, Andreas, would argue yes, in rare diseases you could take it all the way, all the way through. I think it's, you know, cheap enough to do that. And I think that that would be a case-by-case basis. But at least what we're focused in on now, which is cytokine biology, and you know, a lot of the pipeline is immunology and immuno-oncology. It doesn't make sense to take it past phase two. But I will say yes, in rare diseases, like a biotech, you know, could easily take it all the way through.
Yeah, that, that makes sense. Kind of maybe a last one, a bit nitpicky question. So, you know, Absci uses AI and machine learning to basically iteratively perfect binders to a specific target. Is, is this fair to describe the process as generative? I mean, you know, the target itself still needs to be inputted with, with the eventual drug molecule being computer optimized. So you still need to know what the underlying biology before you even start it.
Sorry, Gil, can you repeat that? My Wi-Fi just cut out there.
I'm just wondering if it's fair to describe the technology you guys use as generative, because you still need to know what the underlying target biology is before you start?
Yes, it still is generative in the sense of the antibodies that are generated are not in the training set. So it's generative from an antibody creation standpoint. It's not generative from a biology standpoint. So we're not at the point right now where I can predict what the biology of an antibody is gonna be. But I can use the AI to create an antibody that binds to an epitope of interest and has an affinity that I think will solve my biology. But again, it's not solving the underlying biology itself, but it's a huge accomplishment or like a huge breakthrough just to be able to say, "I have a new novel target.
I don't know what epitope is gonna create the biology, but I can create antibodies that will hit all the surface-exposed epitopes and go test that in the wet lab to find out rapidly which of those gives me the biology." And we've never been able to do that before. So I think it's allowing us to get to the right biology faster and quicker before, and I would argue, in a lot of cases, just actually get to the biology where we never have been able to get to the biology before. Ultimately, we're gonna be able to be at this point where you can use AI to predict the biology, but we're just not there yet.
It goes back to the data, like, predicting biology is super complex, and you have to have scalable data there, and I don't think we have scalable data for predicting biology at the current moment.
I think it's completely fair to say that there's still a very high unmet need when it comes to targeting targets that people know about but haven't been able to target. So that definitely a clear target of pursuit here.
But again, I think that, like, the great thing is like we have two platforms. One is, like, going after these undruggable targets, but then we also have the AI platform on the target discovery side that allows you to get new novel targets as well, and this is kind of where that diversified portfolio comes into play. And again, for your first asset you take into the clinic that's, you know, designed with generative AI, you don't want to be taking both biology risk as well as technology risk. You wanna be able to effectively demonstrate that the technology is working, and that it wasn't the biology that ultimately, you know, failed, even though the AI, you know, created.
We structured this in a way that we can really highlight the power of the AI without taking a ton of risk. Now, on the flip side, we do have, as I said, 301 that is, you know, in our pipeline that we're really excited about, but we're not gonna have that as our lead asset.
All right, Sean, thank you very much. We're at time.