Absci Corporation (ABSI) Earnings Call Transcript & Summary

Good afternoon, and welcome to the 43rd Annual JPMorgan Healthcare Conference. My name is Dave Praharaj, and I'm part of the healthcare investment banking team here at JPMorgan. Today, I have the pleasure of introducing our speaker, Sean McClain, Founder and CEO of Absci. In terms of logistics, please reserve any questions for after the presentation. For those in the audience, the mic will be passed around. And for those viewing on the web, please submit your questions online, and I'll be able to view it through the iPad on the stage. With that, take it away, Sean.

Awesome. Thank you. I'm Sean McClain, the Founder and CEO of Absci, and I have 3 important updates to go over with you today. First is our partnership with AMD, and how we're working with AMD to scale our compute to get better price performance and better training resolution. Second, we'll be going over ABS-101, our potential best-in-class TL1A antibody, and some really exciting new data that we have on that. And additionally, we're going to be going over ABS-201, a really new exciting breakthrough potential therapy in the category of hair regrowth. Now before diving into these exciting updates, let's look at last year. What did we accomplish? We are able to launch a new, exciting de novo model, where we're able to now take a target of interest that has no known binder and generate antibodies to that particular target of interest, being able to go after hard challenging complex targets such as the HIV caldera region. We've been able to successfully apply this model to the execution of partnerships with AstraZeneca and Almirall, looking at GPCRs and ion channels. We've been able to enter into strategic partnerships with AMD, where they've made a $20 million investment in Absci. We've been able to meet our guidance for 2024, entering into 4 new partnerships, which I will talk about on subsequent slides. And we've been able to advance our pipeline. ABS-101 is a potential best-in-class TL1A antibody, which will be in the clinic early this year with a Phase I interim readout in the second half. Additionally, we have ABS-201, a new potential therapy in the category of hair regrowth, going after androgenic alopecia, targeting the prolactin receptor. Additionally, we have ABS-301, a potential novel IO target, and ABS-501, a potential best-in-class HER2 antibody. And what's exciting about this particular antibody is for trastuzumab-resistant cell lines, we see efficacy. Now over the last 5 years, we've been taking our platform, our data generation engine and applying it to AI, this merging of tech and biology. And what's allowed us to have the success that we've had in reflection, it's 4 key ingredients: it's data, it's the models, it's compute and the multilingual expertise. This is what's led to Absci's leadership position in the de novo design of AI with our AI models. Now let's talk about the data. We've always been a data-first company. The company was founded building synthetic biology technology that allowed us to scale protein-protein interactions, how antibodies interact with targets of interest. We were able to go from screening tens of thousands of antibodies to being able to screen millions, and this was right around the time deep learning was taking off transformers in 2018. And it was this idea, if you could take this data with these transformers, you could really go from this paradigm where you're searching for a needle in the haystack to actually being able to create the needle, in our case, an antibody. And this led to the development of a lab in a loop, this iterative process where we're able to go from data in the wet lab to training our models to then being able to validate them. And we do this in a very rapid time period in 6 weeks. This allows us to rapidly iterate on our model designs and architectures and has allowed us to advance our models at a very rapid pace, and it's the reason we're here today. Now let's talk about these leading AI models. We have two. We have our de novo design model, where we're able to design antibodies from scratch. You have a target of interest, you specify the epitope you want the antibody to bind to, and the model is able to design the CDRs that can bind to that particular target and epitope of interest. Now once you have a binder, we have our lead optimization model. This is based on state-of-the-art protein language models. And this is where we're able to co-optimize simultaneously different parameters such as developability, manufacturability or what I like to call smart features, such as pH dependency, binding in the tumor microenvironment, but not binding in healthy tissues. These leading AI models are used to create novel and differentiated therapeutics. We don't want to just use AI to make things faster and cheaper, we want to use AI to go after the hardest problems that still exist in our industry, being able to go after GPCRs, ion channels, these undruggable targets and being able to then apply engineering principles to introduce precise control over the designs so we can enable smart biologics such as pH dependency, being able to increase or enhance potency and MOA. That's why we're building these models, to solve the problems that still exist within our industry and to create these novel and differentiated therapeutics. Now let's dive into 2 case studies that illustrate this. We went over these case studies at our R&D Day in December. And so I encourage you all to click the QR code if you are interested in diving deeper into these, but I'll mention them at a high level. First, on the de novo design. We partnered with Caltech and the Bill & Melinda Gates Foundation to go after HIV. The goal was to design an antibody that could be a neutralizing vaccine for HIV. And the researchers at Caltech, Steve Mayo, Pamela -- and Pamela Bjorkman, discovered the caldera region, this highly conserved region in HIV. Now the issue with drugging this region is that it's a deep crevice. And no traditional technologies have been able to generate antibodies that can bind to this highly conserved region. And we are able to use our AI models, our de novo design model, to generate a very long CDR3 antibody that could bind in that deep crevice. Now this is the first time anybody has ever been able to drug the caldera region. And this is really a great example of how we've been able to apply our models to solve these challenging problems. And not only that, this could be a potential neutralizing vaccine for all different clades or variants of HIV. So if we move to the right-hand side, we were able to show with our lead optimization model that we can start to engineer in these smart features. We showed that we can design molecules that have pH dependency. They bind in the tumor microenvironment that is more acidic, but don't bind in healthy tissues. These are 2 case studies that really illustrate how we are utilizing AI to solve these challenging problems in this industry. It was 3 JPMs ago, we released a pivotal manuscript on de novo design of antibodies. This is where we were able to, for the first time ever, actually anybody in the space, was able to design the HCDR3 of an antibody to bind to a particular target of interest. This was HER2. That's the most -- the HCDR3 is the most variable region in the antibody. And since then, we were able to increase that to 3 CDRs. And now with our latest models, we're to the point where we can design to a target of interest that has no known binder, and not only that, go after challenging and hard targets. I'd like to see -- this very much reminds me of how AGI is progressing. There's different levels, and we're going to continue to progress just like AGI in de novo design of antibodies, and we're really just getting started. And as we start to see these early wins on the board, what we're seeing operationally is that we are spending less money in the wet lab and spending much more money on compute. And that really got us thinking, we need to figure out how to scale compute more effectively. Just last year, we doubled our compute capacity. And that's the reason we decided to partner with AMD, was to get better price performance on these chips, being able to scale much more effectively and additionally, being able to get better training resolution. And I was able to sit down with the CTO of AMD, Mark Papermaster, last week after we closed the deal with AMD to talk a little bit about why we decided to partner with AMD and why they wanted to make us a lighthouse account. And so with that, I'll turn the video on. [Presentation]

So I mentioned that we worked on transferring over 3 workloads or 3 different models. And what we saw from that was pretty remarkable. And it really gets back to why we chose AMD. First, was the unmatched training resolution. With protein design, these protein design models, when you have lower memory capacity, you have to do cropping. You're not able to get the full protein trained on your model. And so you have less context going into the training, which that means that your models aren't going to have as much information and therefore, not be as accurate. But if you can have increased memory, you can get more biological context. There's no longer a cropping of these proteins, and that means that you can actually get higher and more accurate models. And that's exactly what these AMD chips provide. They provide us with industry-leading memory capacity. They have the best memory capacity of any chips in the industry. And so that delivers, point number one, unmatched training resolution. The second is accelerated throughput. Through batch processing, we are able to significantly scale the in silico design and evaluation of our antibodies, which dramatically reduces R&D time and overall costs. And we're able to do this batch processing, again, through the higher memory on the chips. And so this is giving us better price performance as well as better training resolution. Those are the reasons why we decided to go with AMD. And as we continue to scale as an AI drug discovery company, we believe that this is going to be extremely important. Because remember, we're seeing costs go from the wet lab to compute, and we really need to make sure that we can figure out how to get the best training and how to get the best price performance. Just like AMD, we partner across the board with industry-leading companies. And we do this to ultimately get better drugs to patients because we believe that it's not just one company that gets better drugs to patients, it's the whole industry, we do it together. And it's exciting to be able to partner with industry-leading partners like the ones that you see on the slide here. Now the fourth key ingredient to success is this multilingual team. I'd mention that over the last 5 years, we've been integrating biology and AI together. And what I've seen is that your team has to understand each discipline extremely well. Your AI scientists not only have to know how to design leading AI models, they actually have to know the problems that they're solving, which means that they have to be experts in protein engineering. They have to be incredible drug hunters. And the same is true with the wet lab scientists. And we've assembled an incredible team of unlimiters here at Absci that do exactly that. And they're able to take what seems to be impossible and make it a reality every single day. And it's a true honor to be able to work with this team that we've been assembling over the last 5-plus years or so. Now we're taking these leading AI models and applying it to our own internal pipeline that's focused on I&I and oncology. We have ABS-101, which is a potential best-in-class TL1A antibody going after IBD. We have ABS-201, which is a new exciting potential therapy in hair regrowth, essentially common baldness or androgenic alopecia going after the prolactin receptor. We have ABS-301, which is a novel IO target, which we'll be disclosing the in vivo efficacy data earlier this year. And we have ABS-501, which is a potential best-in-class HER2 antibody, where we've been able to show superior efficacy in trastuzumab-resistant cell lines. Now let's dive into the latest data that we have on our TL1A antibody. We've been able to show a really compelling profile in this PD study that we did. We were able to show and confirm target engagement. We were able to show a dose dependency as we increase the dose. We ultimately hit a nice ceiling. And third, at the same dose as the competitor molecules, we see improved target engagement. And we say sustained target engagement as well through day 60. We see this as a very encouraging PD profile and target engagement. And especially as this program enters the clinic, we're really excited to see the Phase I interim readout on that in the second half of this year. ABS-201. I think a lot of investors and a lot of people thought that we are going after atopic dermatitis. No, we are going after hair regrowth. We're really excited about the opportunity here. It's a massive market. There's huge unmet medical need. The target itself is a validated target on both efficacy and safety. And additionally, the development path compared to other indications is relatively fast and cheap. Now let's look at the unmet medical need. 80 million to 90 million Americans suffer from androgenic alopecia, again, just common baldness. And the last therapy that's been approved in androgenic alopecia was in the '90s, 25 years ago. And patients and clinicians are looking for better treatment options. They don't want just slowing of hair loss, they want hair regrowth. And they want safe and minimal side effects, they want a durable and lasting effect and a convenient way to administer the drug. And that's exactly what ABS-101 is hopefully going to deliver on. And so if we look at the mechanism behind the hair growth of ABS-101, it's built on the prolactin receptor. And if you look at the hair growth cycle, you start off with the anagen stage. This is the active growth stage. This is where you have new hair growth. This lasts anywhere from 2 to 6 years based on your genetics. And then prolactin within the scalp pushes the phase from the anagen to the catagen phase where you start to see apoptosis and ultimately, your hair falls out and you start to see the hairline regression. And now by blocking the prolactin receptor, what we see is that the catagen phase gets shunted back into the anagen phase and you start to get active hair growth again. And I'll show you on the next slide here that once you're in the active -- once you're in that active phase, you stay in it for that 2 to 6 years based on your genetics. And so this is a really exciting new mechanism for hair regrowth. And not only are we seeing hair regrowth, but data suggests that we can actually restore potentially hair pigmentation, essentially going from your gray hair to your normal hair color. And we see this as an exciting additional upside. Now let's take a look at the translational model that validates the prolactin receptor target. So there is a monkey study that was recently done, stump-tailed macaques. And these are -- this is a population of monkeys that naturally go bald. And so what you see on the screen here, the images, are the tops of the monkey's head. They look pretty bald in the baseline photo here. And as they go on treatment, as they go on this anti-prolactin receptor antibody, they start to -- you start to see the hair regrowth. And treatment stops after 28 weeks, and the hair continues to grow. And you see hair growth and you see hair durability sustained all the way through 4 years. And so not only are you getting the hair growth during the treatment, post-treatment, you're able to see this sustained and durable hair growth, which, again, shows that the prolactin antibody is, again, pushing the hair follicle back into that anagen phase where you're getting the active hair growth. Now how does this -- how does ABS-201 compared to minoxidil? This -- we performed a study. This is a mouse study where we shaved the mice and we dosed at 2 different doses, ABS-201, and compared it to minoxidil. And you can see from the images as well as the hair score on the right-hand side that we see, with ABS-201, superior efficacy versus 5% topical minoxidil after 21 days. The hair grows much faster and again, seen superior efficacy for ABS-201. Now let's dive into the market. ABS-201 represents a massive market opportunity. As I had mentioned, 80 million to 90 million Americans suffer from androgenic alopecia. From talking with KOLs, there is a strong willingness to self-pay. And the market size, we estimate conservatively is $14 billion, and that's assuming an 11% to 12% conversion rate that you see with Botox. But we think that this could be even much larger. We think that, that conversion rate could be 2x to 3x that, making this market size, obviously a gigantic market opportunity. And that doesn't even include repigmentation, which would be additional upside. 2025 is going to be a really exciting and pivotal year for Absci. We have some exciting catalysts that we're coming up. We have ABS-101, as I mentioned, is going to be entering the clinic here shortly with a Phase I interim readout the second half. We have ABS-201, which is -- which we've just nominated a drug candidate, and we're entering or we're in IND-enabling studies with the plan to enter the clinic early next year. And we have ABS-301, our novel IO target, which is going to have in vivo efficacy data in the first half of this year. And in addition to our preclinical and clinical readouts, we also are guiding to one new large pharma partnership that will be announced this year, which has the opportunity to bring in substantial nondilutive capital. And so you can see that this year, again, we have -- it's a catalyst-rich year with a lot of near-term value inflection points. And with that, we'll open up for Q&A.

I'll kick things off. So you've advanced your AI platform over the last couple of years quite impressively. Do you see room for even further progress? And what does that look like?

Yes, absolutely. As I had mentioned, we see this very much how AGI is progressing. I think they have 6 different stages of AGI or different phases. We see de novo design in a very similar camp. We started off with 1 CDR, 3 CDRs, and now we're designing the whole antibody from scratch going after challenging targets. And so we continue to see that progress each and every year, and I think that there is still a lot of room to continue to improve. One of the areas that we would love to get into is not only predicting the design, but trying to predict the functionality of an antibody, what epitopes should we be going after and which of those are going to give us the functionality we're looking for. So I think those are kind of the next steps going forward in the future. But yes, we're going to continue to progress these models, and I think there's a lot of room left to continue to improve and grow.

And I guess taking a step back, what are some of the key differentiating features and capabilities that your platform has that set it apart from your competitors and enables your drug creation pipeline?

I think Sean is going to let me take one. So I think you saw some of it here today. We are, in my mind, and I used to be an investor in this space, we're clearly the leader in de novo design. So we're now taking these models, and we're designing against targets that you can't address in any other way. I think the HIV case study is a great example of that. But we've done other work with partners against ion channels and very difficult epitopes and transmembrane proteins that we can't share publicly where we've seen success, and we've seen success in a rapid amount of time. In one of those partnerships, we designed against a difficult target de novo. In 6 months, we delivered leads. So we're starting to see the ability to attack these targets that have fundamentally solid biology that can treat diseases where no one else has been able to drug them. And I think that opens up a whole new area of therapeutics that don't exist today. And then on the other side of that, with some of our lead optimization models, you saw Sean talk about designing in pharmacology. And that's a direction that we've -- we started taking early this past year, and you're seeing the results here. Now we're generating pH dependency in our molecules, if that's something we want to engineer into a TPP. So that flexibility to design in pharmacology is also a key way to deliver differentiation and ultimately, to deliver what patients need.

I think it's important to...

Okay. Are we -- sorry.