Absci Corporation (ABSI) Earnings Call Transcript & Summary

Good evening. I'm Sean McClain, the Founder and CEO. Every time I watch that video, I get the chills. And it got me reflecting over the holiday about Absci and our history, where we started 12 years ago. We started 12 years ago with trying to change biomanufacturing, being able to produce antibodies not in CHO cells or mammalian cells, but in E. coli, a very simple organism to ultimately drive down the overall cost and decrease the time it takes us to get into the clinic. And it started in a basement lab in Portland, Oregon. After lots of failed experiments, sleepless nights, we were actually able to show that you could produce an antibody in E. coli. And the amazing part about this and what I've realized is that E. coli is the hero of our story. You all know that generative AI needs data and lots of it, and that's exactly what E. coli gives us. In a single test tube you have billions of E. coli. You can take DNA, put it in and have antibodies, billions of antibodies being produced and screening those with your ACE assay -- with our proprietary ACE assay, allowing you to get billions of protein-protein interaction data points to feed into the model. And so our vision has transformed from biomanufacturing to drug discovery -- and not only drug discovery, but drug creation, changing this paradigm of finding the needle in the haystack to actually creating it, being able to design the biologic the right way the first time, being able to get all of the attributes, the functionality, the manufacturability, developability. And this is going to change how we do drug discovery and biomanufacturing. And I'm excited to talk to you about how we are changing the industry with drug creation today. We will be making forward-looking statements in this presentation. Now there's been a lot of buzz around AI, generative AI. So let's just take a little scroll down history lane. AI started off with classification, being able to identify images: Is this image an apple? Is it an orange? Or is it a human? And now we're in this new phase of AI, generative AI. It's actually creation, where you're able to go from text to image. And this image we have right here was actually created from this large diffusion model, [ Dolly ], where we put in the input -- our Chief Morale Officer, for those of you that don't know, she is my dog. And we had an AI create an image of our Chief Morale Officer in the lab on a computer. And that's the image it created. Now this image never existed. It was created from text. Now what if you could take the same concept of going from text to image but do it for drugs, being able to go from target to antibody at a click of a button. That's what we're doing here at Absci. We're taking our synthetic biology platform along with generative AI to go from drug creation -- or from drug discovery to drug creation. Now why do we want to apply generative AI to drug discovery? Well, if you look at how the process is done, it's completely broken. It takes 5.5 years on average to get a new drug into the clinic with very, very low success rates, 4% or less. And there's a reason for this, because we use biological systems to discover drugs. We have, let's say, phage display or immunization. Let's just take immunization, for example. You take an antigen or a target. You inject it into a mouse. The mouse creates the antibody for you. But you have no control over what the mouse ultimately gives you from an antibody perspective. You can't control what epitope it hits, what affinity it's going to give you, what the manufacturability parameters are going to be, the developability. You have no control over this. So you have this iterative process. And this is why it takes 5.5 years. And you ultimately have to sacrifice one attribute for another, and so you're taking suboptimal hits into the clinic. Again, this is why we have a 5.5-year time to -- from discovery to IND and a very low success rate. So how do we go about applying generative AI to biologic drug discovery? Well, let's just take a look at the landscape. There's been a lot of really exciting companies being formed over the last few years. Most of them have been in small molecules. And there's a reason for that. And I'm going to hit on this point over and over and over. It's all about the data. With small molecules, you have access to that data. Anybody can go screen 1 million member small molecule library and get the data and train their models. But with biologics, it's completely different. Every single antibody you want to screen and test, you have to make in a living organism, unlike making it synthetically by chemist. And the way you produce it is through mammalian cells or CHO cells. And the scalability of that is extremely poor. You can maybe screen 1,000 to 10,000 antibodies in a given week, but that's not enough data to actually train generative AI models. But here at Absci, we've solved that scalability problem, and it all goes back to the basement lab and the hero of our story, E. coli. We engineered E. coli to produce antibodies for the first time ever. And again, because we're in a microbial system, we can build large billion member libraries, and we can take those antibody libraries, those DNA sequences, put it into a test tube of our engineered E. coli and have every single E. coli in that population making a different antibody. So in that single test tube, you now have 1 billion drug candidates you can screen instead of the tens of thousands that mammalian cells would produce. Now you have to actually go and screen those. How do you look at the functionality? This is where we built our breakthrough proprietary assay, which we call our ACE assay, where we're able to interrogate each and every cell, looking at the protein-protein interaction, how tightly is it binding to the target. And all of this data, billions of protein-protein interaction data points is fed into our generative AI model. So how does this all work in practice? It's through our Integrated Drug Creation platform. It's data to train. We're able to screen billions of protein-protein interaction data points that are fed into our AI model and we use that to create. And then ultimately -- and this is a really, really important piece, wet lab to validate. There are so many models out there that don't validate in the wet lab. And we are able to validate 2.8 million unique AI-generated designs a week. I was actually just talking to an AI scientist at one of the large tech companies, and he was telling me that it took over 1.5 years to validate 14 proteins that came out of his model. We can do 2.8 million in a week. So it's not just about the throughput of data. It's also about how quickly we can iterate. We have cycle times where we're able to go from data to train, AI to create and wet lab to validate in a 6-week time period. And this has enabled one of the biggest breakthroughs in the space that we just announced today that I may be talking about here shortly. And it allows us to rapidly iterate on our models, what architecture should we be using, what hyper parameters. And it's dramatically allowing us to decrease the amount of time it takes to get new drugs into the clinic as well as increasing overall success rates. And it's allowing us to attract the top AI scientists in the space. We have scientists that have come from Tesla, from OpenAI, from Facebook. And they're coming here because we're essentially a tech company. They can iterate on their models as fast as they can at a typical tech company because we're able to get the data in a very rapid manner and we're able to train the models and validate them. And this leads me to a very exciting breakthrough that we just announced today. Absci is the first and only company to design and validate new antibodies with zero-shot generative AI. So let's define zero-shot. This means that the model has never seen an antibody that binds the target or a home log. It's very simple. We go from target as the input, and the output from the model is the antibody that binds to the target. We're doing this all from scratch on a computer. No one has ever done this before. This completely eliminates all the biological discovery technologies that currently exist. And this truly is the future. Now we demonstrated this across 4 different targets. And I encourage you to take a look at the preprint or the manuscript that we released yesterday. And the manuscript talks about this working on 4 different targets. It was HER2, VEGF, COVID and an undisclosed target. Now it was a huge accomplishment just to get this to work. But was it actually designing therapeutics that were going to be relevant? Could this actually work in the clinic? Well, I'm here to say that it blew our socks off and we were shocked by the results here. So one of the results I'm going to show you here on the right is a HER2 antigen. So we took a HER2 antigen structure, fed that into our model and we generated antibodies that bound to the same epitope that trastuzumab bound to. And what you see highlighted here on the right is trastuzumab overlaid -- the trastuzumab structure overlaid with the AI-generated antibodies. And what's really interesting here is that the sequences are extremely diverse. One of them is actually 90% different than trastuzumab. But what you'll see is that the structure and the side chains that are necessary for it to bind to the epitope on HER2 are actually conserved. So the model is actually learning what structure is important for binding to that particular epitope, but yet having a huge sequence diversity. This is really exciting. Like this has been a very well-studied antibody, but yet none -- no result has ever come up like this. And it's because we're able to search a much larger search base. So what does that search base actually look like? Here we show the search bases that we looked at, millions, billions, trillions and quadrillions. And when we looked in a search base of quadrillions, we were actually able to have binders or antibodies that bound to HER2, about a dozen of them. That's incredible. Literally, all that went into the model again is the target structure and we got the output of sequences that actually bound. And I want to make it clear. These are all wet lab validated. This is not the model predicting affinities. This is -- or predicted affinities. These are actually measured affinities in the lab. And the really exciting part is not only are the sequence diversities exciting, but the affinities as well are very diverse. You have high binders, you have low binders. And this is, again, really exciting, especially from an IP perspective, because you could essentially look at this and say, "If I had a brand new target, let's say, and I was able to then use this model to design an antibody or antibodies that bound to all the epitopes at various different affinities and go in the wet lab and validate that," you could essentially take out a whole new target from an IP perspective. So not only is it helping us design better therapies faster, but also is a huge advantage from an IP perspective. Now you can say, "These sequences are so diverse, are they going to be immunogenic? Are they actually going to work in the clinic?" Well, we built this naturalness model that basically looked at -- that looks at antibodies and determines how natural they are. And this model we showed in the manuscript 6 months ago that the naturalness is inversely correlated to immunogenicity and correlates very high to developability and manufacturability. And you can see that the model in zero-shot was able to actually design antibodies that actually had higher naturalness than trastuzumab with high sequence diversity. So not only are we able to get diversity and sequence affinity, but we're also able to ensure that we can have high naturalness of the antibodies that are being produced out of this model. This innovation -- this breakthrough is unlocking new and differentiated value drivers. This is going to allow us to design higher potential biologics with increased success rates. Why is this? Because we're now able -- literally at a click of a button able to design antibodies the right way the first time, being able to hit the epitope you want, the affinity, the naturalness, developability. And this is what's ultimately going to increase the success rate. And we're going to be able to dramatically decrease the amount of time it takes to get new drugs into the clinic. And then, additionally, this will start to increase options for personalized medicine, and as I've already talked about, being able to broaden our IP landscape. Now let's dive into accelerating time to clinic. So it currently takes roughly 5.5 years to go from idea to drug in the clinic in the traditional sense. Now with Absci's breakthrough with this AI de novo model being able to design antibodies from scratch on a computer, we're anticipating that it's going to take roughly 18 to 24 months to get a new drug candidate into the clinic. And if you look at the overall cost, we estimate it's $10 million to $15 million to get a new drug into the clinic or a new biologic. And with Absci's platform, we're estimating it will be $5 million to $7 million. So you're able to get more shots on goal, you're able to have antibodies that don't have suboptimal parameters but have the optimal parameters from the get-go and you're able to ultimately increase the overall probability of success on these. And so again, this is a huge game-changer on getting better biologics to patients faster as well as getting them into the clinic at rapid speeds. Now it's not us just tooting our own horns. We have our technology validated through industry-leading partnerships, and one of them I'm going to focus in on today is Merck. We closed a $610 million deal with Merck last year for 3 targets. Additionally, we had developed some bioprocessing enzymes with them as well. And I will say that, that program has gone extremely well. And we're in talks of what targets we're going to be moving forward with. And so not only is this validating from closing the deal, but we've also executed on this program over the last year and it's been an extremely successful partnership for us. So how do we continue to see this drug creation revolution through better biologics faster? Well, it starts off with ensuring that you have an amazing team. We have over 200 Unlimiters. We call ourselves Unlimiters because we turn the impossible into possible every single day. We have a 77,000 square foot campus, which in the video you saw some of that. So state-of-the-art lab. And we also have an office in New York, which is our AI headquarters, along with an innovation center in Zug, Switzerland. We currently have 17 active programs, a $160 million balance sheet, a strong cash balance sheet that gets us greater than 2.5 years of runway. And additionally, we have over 200 patents filed. And again, I'm going to hit on this, the data -- we're able -- in our current facilities able to generate billions of protein-protein interaction data points in a given week to both train models as well as validate. It's all about the right team. What we accomplished today and that manuscript we put out, it would not be possible without the extraordinary team we have. Yes, this is the executive team, but it spans so much more than the executive team. It's about getting people in a company that want to change the industry, that want to merge 2 industries together, biotech and tech, to be able to start designing better biologics through generative AI. And it's an honor for me every single day to work beside these individuals here, and in particular, Penelope, our Chief Morale Officer. Absci is leading the way in drug creation and seeing our vision through of being able to develop therapies at a click of a button. And if we go back to the basement lab, it took us 10 years to develop wet lab, scalable technology to train generative AI models. And we started integrating generative AI 2 years ago, creating our Integrated Drug Creation platform. And that has led to 2 huge breakthroughs. First was the AI lead optimization model, and now it's the AI de novo antibody design, being able to design antibodies from scratch on a computer. No one has ever done this before. This will revolutionize the industry. And just think of where we're going to be in the next 1 to 2 years, seeing our vision through of going fully in silico. So please join us in the drug creation revolution and getting better drugs to patients faster than ever before. All right. Any questions? I don't think our analyst is here. So I will take questions.

Just if you can tell a little bit about the software resources you have put and what would the software look like 3 to 5 years from now in terms of its capabilities?

Yes, absolutely. So where I see this going is being able to -- so right now, we can go from target to antibody hitting the epitope that we want along with affinity, naturalness. Where I want to take -- where we want to take things moving forward is starting to incorporate in the biology. So the biology is not incorporated. So let's say you have a brand new target. You don't know what epitope is going to give you the ultimate biology you want. But right now what we can do is instantaneously generate the antibodies to bind all the epitopes of various different affinities, go into the lab, test that and figure out which of those achieves the biology that you want, and ultimately, feeding that sort of data back into the model so we can be predictive in the future of the actual biology itself, not just predicting the protein-protein interaction. So that's ultimately where I want to be 3 to 5 years from now. And that's what's going to also get us closer to kind of the vision of personalized medicine as well.

You mentioned design make test cycles on around 6-week time scale. Is that right?

Yes.

What sets that time scale? What is the rate limiting step in that?

Right now, the -- I wouldn't say it's the rate limiting step. But what we have to factor into that 6-week time period is DNA synthesis, which is anywhere from 1 to 2 weeks within the 6 weeks, which we have no control over. And then the other 4 weeks is Absci's operational time. And that's where -- we've operationalized both of these extremely well. I think we can continue to shorten those. But the one thing I will mention is that we don't wait for one cycle to be done. We stagger them. So every week, you start a new cycle. So you're not -- it ultimately doesn't become rate limiting.

I have a question. With the -- you mentioned yourself as a technology company. Are you thinking about licensing this technology with multiple partners? That's my question number one.

Yes. So we do not plan on giving the software to our pharma partners to utilize in-house. We still have a partnership business model where they come to us and we use the technology to develop the drug asset and get it to an IND-ready stage. And so we at this point in time do not plan on letting our partners actually use the generative AI models.