...My name is Andrew Pucker. I'm part of the healthcare investment banking team, and I'm joined by Ben Taylor, CFO of Recursion, and Sarah Sherman, head of IR. Ben, thanks for joining us here today. Maybe just to get started, to orient the broader audience, tell us a little bit about just the background of Recursion. There's a lot that we could go into, so maybe talk about the business from the perspective of the technology platform, the clinical pipeline, and maybe just the overall strategy and business model to get us started.
Sure. Yeah, a couple of important points there. So, Recursion has been around for about thirteen years, and the underlying mission of the company has, throughout that period, been to look at why drugs fail and try and find a better way to create a predictive model around that. So whether that's predicting biological connections and reactions that haven't previously been discovered, designing better chemistry, designing better clinical trials, all of those things go into reasons that a good scientific idea may never reach the patient.
And our goal has been to design that in from the very beginning in the drug discovery period. So, our platform has really grown over that period. If you think about each program, we're solving different problems.
Sometimes it is a chemistry problem, sometimes it's a biology problem, sometimes it's a patient selection problem, sometimes it's all of them. And so when we design those models for that specific program, we then integrate it into our unified platform. What that has resulted in is a technology base that has literally thousands of different models, hundreds of algorithms, and about 65 petabytes worth of data behind it.
Almost all of that data is proprietary data that we have either created or had created specifically for ourselves. And so that is a major differentiation for us across this. Not only having the data, which is a very sizable amount, but also knowing how to use it in a better way, because having data on its own wouldn't be enough.
This is one thing that we've learned. When you first start collecting data, it's actually you have to just broadly collect data, and over time, you start to understand what that data actually means, and you can ask more and more sophisticated questions to drive better modeling, to drive better predictions. We've been able to build that into our platform as well, where we have this integrated data architecture that is feeding into models that were built specifically for it,
and have now, over many different programs, been validated on it. What all of that comes back to is really our fundamental mandate. Our mandate is a little bit different than a traditional biotech company, and a lot of our founding investors said, "We actually want you to create a fulsome business model.
We don't want a company with a binary risk profile. We want a company that balances its risk and is able to scale and repeatedly accomplish better quality products. And so that's really what we've been building and continue to build. That is exemplified through our partnerships, where we've brought in almost $500 million to date from partners like Roche and Sanofi, but also through our internal pipeline, where we've got four drugs in the clinic now and some more on the way.
Great. Maybe to build off of that and to double click on one of the aspects of the business is your partnerships. What does Recursion bring to the table in these partnerships? Help investors really think about what is differentiated relative to what large pharma can do by themselves in terms of investments in AI, ML internally, and then also Recursion, maybe versus some of your peer companies that could be vying for the same partnership dollars with a Sanofi, a Bayer, a Regeneron, et cetera. Maybe just paint the picture there, because I think it's an important part of the business.
Yeah, and there's a couple of different pieces to that. In some areas, it is a novel technology that literally no one else just has. So if you look at our partnership with Roche, they paid a $30 million milestone to us last year for a completely novel map of neural cells, where we were able to basically do a gene knockout of neuronal cells across the entire genome, and then are able to use that to map biological function of those neurological cells,
potentially finding completely novel targets, as well as understanding how those cells function in a better way. So that's something that neuronal cells are incredibly difficult to work with in the lab, and that's part of the reason why there are so few neuroscience targets out there in the world today.
So that was an example of Roche coming to us and saying, "We want to leverage the phenotypic platform to try and understand biology in a way that it hasn't been understood before." Sometimes it's for our data. All of that data that we've created, we don't use it as a service. There's no SaaS aspect to what we're doing, but what we can do is then take it and use it to define better targets, to design better drugs, to understand patient populations in a better way. And so what we offer our partners is a composite of that. We can help find new targets that haven't been seen before.
We can help identify which patients are going to benefit from it the most, and then we can help you design a compound that has chemical properties that are difficult or impossible to achieve using traditional methods. And so the last piece that I was talking about there, the chemistry platform is another one where we've already had six programs in-licensed by partners.
Sanofi, in the last eighteen months, has paid us four different milestones for achieving different design milestones. And what we're doing there is using the power of multiparameter optimization in really complex modeling systems, along with generative design of molecules, to actually solve problems in a new way to how traditional molecules are designed. And so that's allowed us to improve the properties of chemistries in a new way.
So you can see how we actually do. I hate to use hypey words, but end-to-end capabilities of -trying to do that, initiation through the validation of the target, now designing, the chemistry to have a much better, more precise profile, and then doing the translational work to help, inform the clinical trial design. We also do clinical trial simulation and patient modeling as well.
Mm-hmm. So it, it definitely sounds like you guys have expanded the, your capabilities with, with partners. Maybe taking a step back, I'd, I'd be very interested, and I'm sure investors would be as well, too, to just get your perspective on overall TechB io. It feels to me like we're still in very early innings, to use the baseball analogy, of, of where we are in terms of the development of AI-based tools and deployment, especially in biopharma, across this broad spectrum of whether it be target discovery to clinical trial optimization.
You probably have a, a better and more informed vantage point than most. What's your view on where we are in the development and deployment of AI in biopharma? I'd be curious to hear, as you look ahead, what do you see as some of the current or potential future developments that could accelerate that? Then a two-part question. Paint for us the picture of, in a future state, where does Recursion fit into that business model in partnership with pharma, as well as just your own in-house clinical drug development?
Yeah. So, it's important to think of AI as a tool, not an end. So sometimes I feel like we're using computers where the rest of the industry may have a calculator to solve a problem, like if we're using AI. And it's not just ourselves. I think AI is that incredibly powerful tool for doing multiparameter optimization, for doing correlation analysis. It's basically a way to execute multiple streams of logic simultaneously to identify something that a human alone wouldn't be able to do. And it's definitely reaching far beyond the capabilities of software because you're integrating in efficiency. I mean, it's sort of like what we recently announced with MIT and Bolt two.
So, what that capability was, is not only looking at protein folding like you had with AlphaFold, but also trying to think about how do chemistries dock in to those proteins. Now, that's traditionally a physics-based modeling question. The problem with physics-based modeling is it's intensely compute, well, it's very compute intensive, so you're going to be holding that until the last to do the analysis, because you don't want to put the cost and time into running that experiment. But what you're able to do by integrating the principles of physics along with AI, is basically you're applying statistical analysis to that compute intensive physics-based modeling. So you're not brute forcing the physics.
You're saying: How can I intelligently do the experiments I need to do to get a good idea without doing all of the experiments? And so what we do is apply that throughout, and that's what AI allows you to do, is apply that throughout the different experiments, designs, analysis that we're doing. So it's far more efficient. Now, the difficulty, it's not actually hard to design an algorithm.
Lots of people can do it. What's really hard to do is design an algorithm that not only works, but you know it works. And that's what you need to do when you're creating a system that has thousands of integrated models, right? Like, you need to know each one of those models is independently working or how they're going to work together. And so that comes down to a couple things.
Quality of data, I think, has been really highlighted to people throughout the industry. If you don't have good quality data, then you're not going to get good quality results. That may be fine if you're doing, you know, an internet search, but it's not fine if you're trying to decide where to put an atom on a molecule. And so, we've focused a lot of time on creating that data set I was talking about earlier. The other part is you have to understand what good looks like. You have to understand how to validate it. And that's where we do two different techniques. One is everything that we do virtually, we try and have an experimental technology that's paired with it.
So we actually have significant internal laboratories that we're able to do, you know, very precise cutting-edge experimentation to say, is this actually predicting what we thought it was going to be predicting? And if it is, or if it's not, feed that back into the model to make it better. And so, that's a really critical part. You have to have some sort of, experimental validation and/or, to have multiple models that would, contradict or reinforce each other. And so this is where you've seen us expand into transcriptomics, and you've seen us bring in real-world data and different data sets. So let's say we see a signal in our Phenomics platform, which is basically how does this cell work? How is this cell reacting?
Signal from that could be right or could be wrong, but if I can compare it to real-world patient populations and the data that was done through clinical trials, and I see that same signal, I'm probably onto something. Or if I'm seeing that same signal come out of a completely independent transcriptomics experiment from what I did in the cellular environment.
And so we have, we're constantly working on integrating all of this together. So for an industry perspective, what we always look for is... Not only is the technology useful, most of them are, because, you know, there are so many points to improve inside of our sector, but also how is it validated? What's the use case behind it? And, how robust is it in being able to do new things in the future?
In that, I think, there is some good work going out there. Most of the industry is focused on point solutions, though, where they're sort of tackling one of the problems. And we've taken a little bit different approach, largely because we've been around a lot longer. We've just had time to grow out that platform and do different parts of it. But we really want to see that integrated technology come in and say, how do all of these things interacting together get me to a better answer?
Mm-hmm. And so it sounds like the integration or the integrated broader approach is a source of, is a real competitive moat as you think about what-
Absolutely.
is able to do versus pharma. And maybe talk to me about, you mentioned the work that you're doing with MIT. My understanding is that's open source. How should investors think about where you decide things ought to be open and transparent in terms of the models that you're building versus what you're going to keep proprietary?
Yeah.
Also related to that, how should investors think about the encroachment, whether you see any at all near term, from really tech native AI companies into the new healthcare sphere where you operate today?
Yeah. So, on the first part, in trying to decide what we would make open source, because Boltz Two wasn't the first thing. We've released some of the different data sets out publicly. I think what we're trying to do is understand where can the community benefit from and also educate us as well. Like, if we look at something like protein folding scenario we actually expect will likely be commoditized in the future. And so we would rather be on the leading edge of helping that happen, and getting that out there and helping you know, giving people tools because there were some of the tools were there, were locked behind different business models or SaaS models.
And so we wanted to put something out there that would help people get access to it and build on it. And that, I think, helps the entire community advance on those elements. So sort of speeding up a process that we believed would happen over the next probably 12 or 18 months anyways. But that's not really where we drive our value. If you think about the value that we contribute, it's when we're doing a program, it's the specificity of the data, the accuracy of the data, it's the integration of the models and the workflows, and how do you get it all to fit together.
And so actually, that idea of trying to understand the protein structure and maybe some of the basic ligand binding is really the first step out of, you know, a hundred, or sometimes it feels like a thousand. And the more tools we have in that space, actually, it just will benefit us. So the more complicated design elements, the more complicated biology elements come from the integration of all of the things going on, not the, "Hey, I'm going to understand you know this one protein structure or this one binding aspect.
Mm-hmm. Is that what you think has been one of the primary separators of Recursion versus some of your peers that may not exist anymore today? I feel like there was a fair bit of investor excitement several years ago around tech bio as Recursion and others went public. And as you look at the landscape today, it's very different business models, but it's Recursion and Schrodinger, and many of your peers, both public, obviously, you guys are the beneficiary of the merger with Exscientia.
Yeah.
But a number of notable public and private, in particular, peers are no longer around, at least in their original state or business model. Is it-- How much of that is just natural first mover advantage versus the strategic approach that you guys have taken around data and integration? I'd love just your perspective on-
Yeah. Oh, absolutely.
on how that's evolved.
It's definitely a mix. I think I think one of the differentiators is whether you're looking at legacy Recursion or legacy Exscientia, both companies were really focused on: How do I put this together in a patient-first environment? Like, how do I think about where it needs to be in the clinic first, and then I'm just going to assemble whatever technologies and science I need to try and get me there.
And that's very different from I've got a technological innovation that I want to find a use for, right? And so I think that helped guide us towards things that we're going to have more, more use cases, more practical application. So that has certainly helped. There's always a bit of luck in that as well. I think we've had fantastic partnerships.
It's hard to underestimate the value of that, because not only do they bring in great expertise and capital, but they also keep you honest, right? Like, a partner is not going to give you hundreds of millions of dollars for nothing, right? Like, they want results. We have to, you know, meet with them, you know, monthly, quarterly, to review how the programs are going, and if they're not going well, they're not going to pay for them, right? And so you've got to continuously produce it.
And so both companies, and now together, we really have been able to keep that as something that has always made us ask, "Is this system not only sounding like it's going to create value to us, but does our partner actually understand why this would be useful or not? Or are the results actually compelling to an independent third party?
Mm-hmm.
And so that's been hugely helpful. But then the focus on data, I can't underline it more, because what that has always guided us towards is one, experimentation to be able to validate our models. If you're always reliant on a third party, let's say a CRO, to do your experiment, you're always reliant on their capabilities, their output, whatever data that you ask for. It's an inherently biased experiment as most experiments are, but it's even more so when you're doing it through sort of a third party.
So by building that up internally, we were able to not only design the experiments in a way that was more customized for our set, but like if you look at the Phenomics platform, really just do it at a high scale that allowed us to take an inductive approach to understanding where the science should be leading us. Not saying, "Hey, I want to ask this question, CRO, can you go run this test for me," but I'm actually going to create a data set, and then I'm going to let that data tell me where I need to go. That is a very different approach as well. I think all of those combined together. I gotta tell you, I'm cheering for all of the companies in this space.
We're not fighting, or we shouldn't be fighting against each other. I mean, we're in an industry where the failure rates are north of 90% in the clinic. I mean, if you add in drug discovery, you're at, like, probably 95%-97% failure rates.
Like, we need to improve that for patients. That's, that's why I got into this industry. Like, we need to have better medicines that are coming out. We can't do the scientific experimentation that we need to, because when you're in a 95% failure rate, you limit the variables, you focus on one or two things, you invest everything behind one idea.
Mm-hmm.
And so we need to change that. That's what we're competing against. That is, the big bear in the room that needs to be tackled.
Yeah. You mentioned a couple of times, partners and how they validate what you're doing with your strategy and your platform. And I'm also just thinking about the competitive landscape here. Recursion's been incredibly successful in terms of very meaningful partnerships and integrations with some of the largest blue chip pharmas in the world.
How should investors think about the business development forward with existing partnerships, as well as Recursion's ability or desire to partner with other large, you know, the finite number of large caps, as well as a much larger universe of mid to small caps, in the biotech landscape. What are your thoughts on, you know, tell us a little bit about the business development strategy focused on partnerships going forward.
Yeah, and I mean, I keep most of my comments focused on the near term. I think the long term could have a lot of different business models behind it, but in the short term, we've got fantastic partnerships. Obviously, Sanofi and Roche are the largest two that we have, but those have been going extremely well, and we're advancing multiple programs with both and intend to continue doing that and really keep investing and going deeper with those partners. We are always open to new partnerships as well. I think one of the things that I, if I put my CFO hat on and think about is, I don't want to take away resources that we could be investing in our really good partnerships to, you know, start a new partnership.
We'd have to be able to do them both, which we have the capability to do, depending on what that new partnership would look like. But it just puts a higher bar on it. 'Cause we've got room to grow into our existing partnerships. We have a lot of room for expansion in there, and we wanna, we wanna keep driving good relationships into even better places. I mean,
Sanofi has been at the forefront of talking about where they want AI to take the industry, and they want to be a real leader in it. Roche has been incredibly innovative on how they're thinking about new target ID and what neuroscience could look like in the future. And so we wanna, we wanna continue investing behind that. But we are always thinking about new partnerships.
We do also think about, you know, the right long-term homes for our internal pipeline. They may be with us. We may take everything through to commercialization, or we may out-license it at different periods to partners. I think, there's nothing that is off the table from our perspective. It's all a matter of what is gonna make the right sense for that specific program at that time, and the nice thing about not having a lead program in that traditional biotech sense is we actually feel completely comfortable doing this, that with any of the programs.
Mm-hmm.
So I think that's where we see the partnership continuing to go. I wouldn't expect us to do a service-based or a SaaS model anytime soon, largely because it would require such a simplification of what we do. And that 95% failure rate I was talking about earlier suggests how complicated this problem was. If it was something that you could simplify, somebody would have done it already.
And so until we see those probability of success numbers really start to move, I think making sure that we have full control over how the platform continues to grow and evolve and is used is important. If we did see that in the future, that's why I'm saying long term could be a lot of different potential aspects.
I mean, a couple of years ago, a lot of our internal UI didn't exist, and now all of a sudden, you know, we've greatly increased our ability to be more efficient in our internal things, right? That'll keep going, and as we become more efficient internally, that could eventually translate into something that's very easy to use externally, too. But, in the very near term, we love our partnerships. We really like the structures that we have. We've got terrific economics that come out of it. And so, probably more of that.
Okay. So we've talked a fair bit about partnerships, technology platform. I'd love to dive into the pipeline, the clinical pipeline.
Mm-hmm. Yeah.
You came through the merger with Exscientia, a very rich pipeline that was really refocused earlier this year. Maybe take us through the pipeline and some of the key programs. There's no lead program, which is a little bit unique here. So give us a little bit more detail there, and then maybe just point us in the direction of what are the meaningful near-term clinical catalysts that investors should be focused on?
Yeah. Well, and it's interesting. All of our pipeline programs have actually two points of important validation. One is, did the platform do what it was supposed to do? Like, for example, if you look at our CDK7 program, for example, the data that we announced back in December of last year. When we looked at the PK/PD modeling, when we looked at some of those different things,
they were right on where we had predicted they should be and where we wanted them to be for human biology. So that's one part of it, saying, "Yes, design that hadn't been accomplished previously has been accomplished to the extent that we know." The second question is, does that then translate into a good potential commercial product?
And so then you're looking at more of how does the clinical data read out, you know, in a more traditional sense. And so the exciting thing is, over the next couple of quarters, we have both coming up. So, there's a lot of people that are thinking about our MEK1/2 inhibitor in FAP. This is one where the phenotypic platform found a connection that no one had seen before. And it is an incredibly high unmet need patient population that are repeatedly going in for scopes and polyps removals and resections because it, left untreated, it will progress on to a cancerous state.
And we saw some really exciting, but very small N results that we released a couple of months ago, showing pronounced effects in the different aspects that are relevant for that. So reduction in polyps, reduction in size of polyps, and an effect on dysplasia. So that's basically the cell looking less cancerous overall, which is really exciting because those are the things that would trigger a patient to have to go in to have surgery and out of a resection or potentially indicate progression onto cancer.
So we wanna see more of that. We also want to see, hopefully, patients continue to maintain some of those benefits with the treatment holiday. So the trial was designed to have patients on for twelve weeks and then take twelve weeks off.
And so some of the data that we presented earlier was that twelve weeks on, what we'd like to see is if patients can go on a treatment holiday and maintain some of those benefits. It actually dramatically expands the patient population, because then you could be looking at more mild and moderate FAP patient populations. If not, then you'll still be able to address the more severe patient populations. But it would be really exciting if we see some of that. So, there's a lot of focus on that coming before year-end. We will have more data on the CDK7 program in monotherapy before year-end, but I think the more exciting data in CDK7 is really that'll come in the combination study, which has started.
CDK7 is a target that, on its own, isn't expected to produce a lot of efficacy. We were thrilled to see some in our initial readout, but it was definitely not expected. This is what's called a cytotoxic or a cytostatic mechanism, so it's not necessarily going to shrink the tumor on its own. But in combination, you should absolutely be seeing those benefits come through. And so, this is one where we've already seen good evidence that the design platform did what it was supposed to do, and so that gives us a real chance where other people haven't had a chance to show that now we can adjust the biology to have an impact on cancer. And so, that would be very significant data if positive.
You know, the CDK4/6 market is a $9 billion market, and we believe that CDK7 has broader application, if it were to go ahead. So that's a very exciting one, a little bit more downstream. Near term, we also have MALT1 and RBM39, which, RBM39, we've given guidance that we'll give initial data first half of next year. This is another one where the platform, the phenotypic platform, came up with a connection that just... it wasn't in the literature. There wasn't anything that you would have been able to find connecting RBM39 and CDK12, which is a really exciting but really difficult oncology target. And you can see it in the cellular biology when you use our platform.
And since then, we've gone on to release additional data showing how it's connected. Some of the third parties have as well. So that is currently in dose escalation. And we've done a more enriched patient population. So we focused on specific biomarkers like MSI high, where patients should be more responsive to that sort of a therapy. So that'll be an exciting readout. If there is any activity, it would just be a completely novel finding showing that Phenomics was able to discover something that has not been out there. You wouldn't have been able to use an LLM or any sort of literature-based model to find it, because there wasn't any literature in existence.
Those are exciting, and MALT1 is a more traditional design story of a drug class that has a known side effect that could really limit its use. We hope that we've designed out that side effect, but we'll know pretty quickly when we are doing the dose escalation work and so are looking to put out more data on that next year.
Okay. How do you internally think about this, and how would you guide investors to think about the platform value at Recursion? It strikes me there's incredible value in the technology platform. You're pursuing different approaches, business models, how to monetize that versus partnerships versus clinical pipeline. The question I have, I'd love to get your view on, is what is more validating or how do you think of it. It's probably both, right?
How do you think about the power and the importance of, to your earlier point, blue chip, large cap pharma companies furthering their partnerships with you, paying milestones, taking on more programs, versus what you're doing in your own clinical pipeline, which has leveraged different components of data and models, et cetera. How should investors think about that? It strikes me as pretty important.
Yeah, absolutely, so this goes back to maintaining a risk-diversified model, so when we did our post-merger strategic review that we basically announced the end results of in May, and we talked about how we were prioritizing different assets and different technologies and really focusing in the company on certain impact areas. It was all around how can we create the best profile to demonstrate that what we're doing makes a difference, and also keep that risk diversification so that we can, we never become reliant on any one program to be successful, and we've got multiple routes to be successful, and so you're right. I would say it's all of the above in the sense of they're all important.
But what it really comes back to is, do you see the arrows lining up in the right direction, right? This is not a problem that's going to get solved overnight. It's not like you flip a switch and all of a sudden drug discovery is solved and that goes from 95% failure rate to none. This is a street fight with, you know, biology and chemistry and all of the unknowns that cause disease and lead to different patient outcomes. And so what we wanted to do is say, okay, we know the different components that go into it. Let's set up a pipeline and set of partnerships that allow us to be economically stable while continuing to demonstrate not only things that show the platform is working, but have their own intrinsic value.
So if you look at any of our pipeline programs, all of them are targeting really important patient opportunities. If you look at our partnerships, we are going into places that no one has gone before. We don't do any me-too work. We don't do any, "Hey, we're gonna just make this a little bit better for a market that's already saturated."
We want to go and design drugs, find biology, meet patient needs, where people have failed repeatedly or they don't even have an idea. And so, each one of our programs takes different components of that and, tries to put it into, rather than solving it all at once, we're going to solve this problem or that problem or maybe these couple of problems and do it in a good commercial market, and we'll just keep driving forward on that.
Great. Thank you, Ben. Appreciate you spending the time to speak with us today, and that'll conclude today's session.
Great. Thanks for having us. Bye-bye.