Thanks for joining the session with Recursion. My name is Alec Stranahan. I'm Vice President and Senior Biotech Analyst covering Recursion, here at B of A. I'm pleased to be joined by Michael Secora, Chief Financial Officer of Recursion. Thanks for being here, Michael.
Well, thank you for having me, Alec. Great to be part of this conference. Great to chat here with you today.
Yeah, great. Fantastic. So, you know, it's a fireside. We'll run through questions. I have a few here, but I welcome anyone from the audience who has a question just to raise your hand, and someone will be around with a microphone. But, you know, Michael, maybe just starting high level on the platform. You know, I think at this point we're all familiar with AI in healthcare, at least at some level. Could you maybe outline the promise and challenges of applying AI to healthcare and where maybe Recursion fits into the equation?
Yeah, happy to. Well, let's unpack that. Let's first kind of talk a little bit about that promise. And I think that, you know, we find ourselves at a unique time and place where there is this confluence of technologies across compute, across big data, ways to control biology, ways to control chemistry. And those integrated together can be utilized by these tools like AI and ML. And they are meant to have an impact on the fundamental problem that we are trying to grapple with, which is around the time and cost to bring a new drug to market. But not just those kind of efficiencies around time and cost, but also understanding biology more completely, understanding novel insights that can be turned into targets and to new kind of chemical scaffolds, and driving increasingly to find the right drug for the right patient at the right time.
To your point, Alec, we can, you know, we can ask the question, where within healthcare could these tools of AI and ML be applied? I would argue anywhere. Anywhere where we are systematically collecting large-scale relatable data are places where these tools could be applied. I would even look at the paper that the FDA put out about this time last year around the uses of AI and ML in drug discovery and development from novel target identification all the way through next-generation manufacturing. But even in that promise, we also see a little bit of what that challenge is. That challenge is the availability of very large-scale connected relatable data sets for which these tools can be applied. I think that only recently have we started to kind of really grapple with how do we curate and generate such large-scale data sets.
I think Recursion, to kind of get into where we fit within this equation, since our founding about 10 years ago or so, we have been very much fixated on the curation and generation of large-scale data sets for ourselves from our own wet labs. We've generated enormous amounts of phenomic data, transcriptomic data, ADME data, in vivo data. We've been able to use our computational tools to map out chemoproteomic reactions, as well as look at the mining of the body of literature using LLMs to understand what does science believe to be true around certain relationships.
We've also partnered with groups like Tempus and Helix to be able to access multimodal patient-centric data sets and in so doing have a more complete operating system to drive novel insights across biological targets, compounds, biomarkers, patient populations in service to trying to move drug discovery and development increasingly to that of a search problem to be able to bring the right drug to the right patient at the right time.
Right. Great. That's, that's a great introduction. You know, when you think about how you're approaching this huge issue, right, this huge problem, which is developing better drugs faster, getting them to the patients who need them, there's a lot of different ways that you can approach this, both in terms of building out your own capabilities, but also monetizing those capabilities, right? You know, maybe, maybe could you walk us through your approach to creating value from the platform today and also, you know, how you're looking at this over the long run?
Sure. Absolutely. Well, if you look at Recursion's operating system, we have taken a lot of lessons and learnings from the technology space. And it begins with that operating system, which allows us to systematically and methodically generate vast amounts of data for which we're able to extract insights and construct models. With those models, we make predictions. With those predictions, we drive those predictions into validation and optimization, and from there into a potential program that we take into the clinic ourselves or with a partner. And that arc, you know, that arc is both applied to how we think about our own internal pipeline as well as our partnering opportunities. And so that gets a little bit into our business model or how we drive value. And there's a three-pronged business model that we have.
One is our internal pipeline, which is focused on precision oncology and rare disease. There is our partnerships, particularly with large pharma companies, which focuses more on large, complex therapeutic areas like neuroscience. And then there is our third pillar around data strategy and how we think about providing access to some of the data and some of the tools that we've been able to generate for potential partners, whether they be on the large pharma side or on the tech side.
Mm-hmm. Great. You know, I've been to your office, and it's a pretty interesting structure. You've got, you know, all the large monitors with all the data scientists and all the programmers. And on the other side, you've got sterile white rooms, you know? How are you sort of balancing developing your machine learning AI with investing in wet lab innovations?
Absolutely. Well, you know, I'm very glad you've been able to kind of, you know, visit us. I think a couple of comments I would make just about the layout of our office is that you'll find an open office plan for which you have computational folks and technologists sitting next to biologists, chemists. There is this integration, not just in technology, but the integration of talent and what that means for creating this novel kind of culture where we have relatively equal parts of life scientists and technologists at the company. And also, to your point about the being able, no matter where you sit, to visibly see the laboratories, that is by design so that no matter whether you're a biologist, chemist, machine learning scientist, you can see what we're doing. It's there. It's visible.
It's always before you to see what work, what work are you doing in service to the automated workflows and the processing of an insight to a program. Now, to your question around the balance between wet lab and computational means, really, this goes very much in line with the company's demographic, that data and compute go hand in hand, and that we relatively spend on equal parts for these efforts. And it's, you know, compute requires data to be able to extract out insights. And as those insights are found, more data is required to understand the greater connectivity to the system, for which then more compute is needed, and back and forth and back and forth. And it is just this connection between these two pieces.
I think even as I've kind of framed a little bit of compute, data, and talent, as far as I can tell, these are the three very important resources at play in this, micro, sector of AI-enabled healthcare. It is access to large-scale compute. It is access to large-scale relatable data sets. And it is access to talent who are capable of utilizing those first two resources. And I think, you know, we've been, we've done very well to try to, have influence on those three pieces.
That's great. And, you know, you've covered, you know, what you're doing, what you're building out today. And, you know, you've got your clinical studies ongoing, which, I want to drill down. But, you know, before we get too deep in the weeds, and this is a question I get all the time, but when do you think AI-enabled drug discovery has its ChatGPT moment? In other words, where does the promise we've talked about become tangible reality?
Yeah, great question. Great question. This is something I have thought a good bit about and have certainly gotten this question as well. I think one thing that I would first acknowledge is that, you know, as a scientist and an investor before moving to the operating side, it appears as if incremental changes often go unnoticed. I believe that is true in science, and I believe that's true in technology, and I believe that's true in trading. I think, and I also believe it's particularly true when one is operating within a system or studying a system that is evolving very rapidly, like AI-enabled healthcare. To get to your point, Alex, about ChatGPT, let's unpack that because it's a great example of, I believe, incremental changes not, you know, going unnoticed for a while. ChatGPT, you know, GPT-1 went live in 2018.
GPT-2 went live in 2019. GPT-3, for which ChatGPT is based, went live June of 2020, about 2.5 years before the big moment, right? And so that technology was already there. It was already many people were already seeing the value. But it was in that coupling of the web wrapper that enabled this pervasive notoriety where it became obvious to all. But many people had already seen the value. And I believe history is repeating itself a bit within the space of AI-enabled healthcare, for which, you know, we are seeing things like last week or so, AlphaFold 3 comes out. That is incredible. You look at us as an industry finding novel targets, designing novel molecules, finding new ways to think about clinical data.
You see life science companies being founded today, really being founded, I think, on the principles around digital and data nativeness, like Xaira, right, which tremendous amount of funding that got announced, and then also some incredible folks like Scott Gottlieb, Jennifer Doudna, other folks being around a company steeped in this idea. And so I just really believe, you asked the question, when do we see it? I believe we're there already. Now, when do we have that aha moment where it's obvious to all? I'm not sure. But what I am confident about is that it's not an if question. It's really a when question. Does everyone come on board? Because it seems to me that these ideas are only going to be more adopted into this space. And I think 10 or so years from now, a lot of this is going to look self-evident.
Yeah. Very good. And, you know, I think you mentioned on your 1Q call that, you know, you're happy to be the trailblazer here, but you're also happy to invite, you know, company on the journey from other, other large companies that we've seen, coming into the fold. So, well, maybe, Michael, one piece that you mentioned, in terms of how you're driving value from the platform is through partnerships. Maybe we can start there. You know, how are partnerships, you know, one way, that you're leveraging the platform? And, you know, maybe if you could just run through the partnerships you have today, either with large pharma or with tech.
Sure. Absolutely. Well, I think we've been very fortunate to have some great partners be around us, both on the large pharma side as well as the technology side. On the large pharma side, here, we have been able to apply our platform more plastically in areas of large, complex, you know, therapeutic areas like neuroscience. And then, to complement that, on the technology side, we have been able to access capabilities and tool and, and tools and data in order to allow our operating system to function more fully. And that is partnerships on the life science side with Genentech, Bayer, and on the technology side, partners like NVIDIA, Tempus, Helix, and Enamine.
Mm-hmm. Great. You know, how are these collaborations going? How much more, I guess, dry powder from the pipeline for additional partnerships? And, you know, where, where could we sort of see, you know, these current partnerships sort of kicking into gear in terms of data?
Sure. Sure. Absolutely. So to walk through some of these and to kind of walk through some of the recent milestones we've had for each. On the large pharma side, we have a partnership with Genentech, in the space of neuroscience and one GI oncology indication. And last year, they have already optioned the first program in GI oncology. And as we continue to do work with them, particularly in map construction, watch on the near term for potential options, perhaps options related to map-building milestones. With respect to Bayer, we've been working well with Bayer. We are operating in the space of undruggable oncology, and that being an update from previously working with them in fibrosis. And as they had changed their focus, we changed with them.
I think in that partnership, that allowed us to actually in-license the most advanced program we had in fibrosis in very favorable terms. That program already in preclinical, the preclinical stage and already going through IND-enabling studies on its way to an IND. That's for a novel target, again, in the space of fibrosis. If we move then to the technology side, I think we've had a great partnership for some time now with NVIDIA. We've been working with them for about four or five years. And they've really, you know, helped us gain access to compute, large-scale, high-performance compute.
Just recently, you know, yesterday and in our earnings, we talked about BioHive-2, which is our next-generation supercomputer, that going live and that being profiled, that being benchmarked where we see ourselves being a top 50 most powerful supercomputer across any industry and the most powerful supercomputer in all of biopharma. And that goes to that data question. And when you command one of the largest data sets of its kind on Earth, you need substantial compute to meet that. And I think, you know, they've been a great partner. And I think even the speed with which BioHive-2 was put together.
Also on the technology side, partnerships with Tempus and Helix give us access to multimodal, patient-centric data for which we're able to train AI causal AI models, also helping us to identify biomarkers, also helping us to target and enroll patients in our clinical trials. And that's already underway with some of the programs that we have in our pipeline. And then lastly, with respect to Enamine, that's a partnership based around chemoinformatics and chemical synthesis. And some of the work that we've done with them, also helped by the work or the collaboration we had with NVIDIA, we were able to execute, I think, an extraordinary digital chemistry calculation looking at profiling Enamine's 36 billion compound library, the Enamine REAL Space, against the entirety of the human proteome.
So constructing this chemoproteomic map to understand ligand-protein interactions and where you might have on-target, off-target effects. All of that helps to more fully, helps to more fully enable the operating system. Now, to your question around where might we have additional appetite, I think there's, there's a lot of, there's a lot of opportunity for us here. I think on the large pharma side, we would be happy to, you know, strike perhaps additional partnerships in other large therapeutic areas like cardiovascular metabolism, among other areas. And on the technology side, I think we would certainly be open to exploring access to other kinds of data so long as it is of scale and relatable and can fit into the data that we've already generated ourselves or, or accumulated.
Of course, there's other things around capabilities, maybe even other kind of modalities that we could also consider in the future.
Yeah. And you mentioned, you know, the GPUs that you got from NVIDIA. That's immediately tangible value. I guess, you know, those were the GPUs that went into BioHive-2. How does having that much compute power set you apart from others in the industry? And does that create a moat for you guys that you can leverage moving forward?
Great question. I think it absolutely is a differentiator, knowing that not just in the tech or AI-enabled drug discovery space, but in all of pharma, you command one of the largest computational resources, but that being needed when you have such a data set of that scale. And it's how we are finding insights that are not known or not well known in the corpus of scientific literature. As a scientist by background, I find that incredibly exciting, what we're able to identify, what we're able to drug, what we're able to take forward. And I think it's a resource that really helps us be broadly partnerable, to many folks, helps us design better drugs, helps us identify biomarkers and get these, you know, and get to the clinic, more quickly.
You know, but I also think that, you know, the compute that we have, I mean, it is an absolute necessity. Neither data nor compute is essential by itself. Both are needed. And I think even as what we framed a little bit in our learnings call last week, we talked about how we are now adopting active learning approaches. So what I view within this space as a major differentiator is data. Access to large-scale, relatable data is a major, major differentiator. And something that we started to kind of frame is the way that we can potentially more efficiently construct maps of biology across multiple cell types and not just with a phenomic basis, but in a transcriptomic basis. We're moving towards having our first genome-wide transcriptomic map.
You know, the ability to effectively derive the greatest amount of knowledge from these maps with the least number of experiments is about how do you not just command a large data set, but how do you set yourself up to continue to add more and more layers of a data moat?
Right. It all feeds into itself. And that's where you get the flywheel that you guys have talked about. Okay. So we touched on partnerships. That's one sort of arm of driving value from the pipe, the platform. You also have your own internal pipeline. And I want to focus on a, a couple of the, the later stage assets, maybe starting with REC-994. This is your CCM asset. It's a superoxide scavenger. Could you maybe talk about the need in this indication and, you know, how have your FDA interactions been in terms of, you know, designing your study and the path forward?
Sure. Happy to. So CCM is a rare disease. It is a very large rare disease. It affects approximately 360,000 patients in the US and EU5. It is a disease of the vasculature that's characterized by lesions that can appear within the brain or in the spinal cord that can give rise to seizures, that can give rise to hemorrhages, that can give rise to other neurological deficits. Now, this disease is caused by loss of function of the CCM 1, 2, or 3 genes. There is no approved therapy. Surgery may be attempted. But given that you have lesions in the brain, sometimes in the elegant structures of the brain, sometimes surgery is not tractable. I think even just highlighting that, the scale of this disease opportunity, yet it is often not a very well-known opportunity.
I think highlights sometimes the vicious cycle that comes with a rare disease with no therapy, perhaps underdiagnosed, and so on. Now, in our conversations in interactions with the FDA, I think that, you know, the interactions with the FDA have gone very well. I think it's important to call out here that we are the first therapeutic candidate advanced in an industry-sponsored phase 2 trial for this. And in that pioneering nature, we have, you know, in collaboration with the FDA, we are measuring about 10 different efficacy endpoints, those being some like looking at MRI imaging around the number of lesions, the size of lesions, the rate of change of lesions, also looking at clinician-measured outcomes, also looking at patient-reported outcomes, as well as the impact on acute stroke.
And so there's a host of these measurements that we are looking at, you know, in collaboration with the FDA and advancing this trial forward, which we're going to have data very soon in this phase 2. We're going to have data next quarter.
Right. Right. And that actually feeds into my next question. You, you started the phase 2, in 1Q 2022, fully enrolled June of last year, data on track for 3Q 2024. This will be sort of the next look we get at clinical data from the pipeline. Any additional framing you can provide around the scope of the data we should expect, and maybe what the next steps could look like, after the, after the 3Q update?
Sure. Well, as I said, there's about 10 different efficacy measures that we're looking at for this disease. We believe that movement in any one, in one or more of these measurements, will give us and KOLs' confidence that a next step for this program is warranted. I think what's also important is that we believe that a combination of certain measurements, such as perhaps something on the more objective side, like the imaging, coupled with something on the more subjective side, like a patient-reported outcome, also gives very compelling evidence to be discussed with the FDA. I also think what's important is that as we ran the phase one trial and that those results out in a paper that one can look at, we saw an incredibly favorable safety profile for that drug.
As we've been running this phase two, as patients have been on drug for 12 months, the vast majority of patients have chosen to opt into the long-term extension study. I think that continues to highlight a very safe molecule for which there could be chronic dosing. I think taking all of that together across efficacy and safety, I think certainly we look forward to interacting with the FDA. I think all of that can help inform a path for registration.
Oh, great. Great. Yeah, definitely looking forward to that update in 3Q. Maybe moving on to the next asset, which is REC-2282. This is for NF2, meningiomas. It's an HDAC inhibitor. You know, could you maybe talk a little bit about why you went after NF2, you know, this, this specific disease? And then we can get into, you know, sort of the status of the program.
Absolutely. So maybe I'll talk a little bit about the indication. We could talk a little bit of why we went after that disease. With respect to NF2, this indication, also a rare disease, affects approximately 33,000 patients in the U.S. and EU5. It is a tumor syndrome that results in meningiomas, intracranial meningiomas, as well as vestibular schwannomas, and that can then lead to loss of hearing, loss of mobility, as well as other neurological deficits. Like CCM, it too has a genetic driver. It is loss of the NF2 tumor gene, and it's a tumor suppressor gene. Here again, there is no approved therapy for which, you know, surgery may be attempted. But again, we are talking about doing brain surgery again.
Why we chose to go after NF2, I think number 1 was the unmet medical need that exists in this disease. Number two, I think that also comes from when we had been applying our platform increasingly to looking at rare disease, particularly rare disease with genetic drivers for which there will be a causal morphological effect that could be picked up in the phenomic data that we had. All of that gave support to this opportunity. And as we carried out preclinical research, we gained greater confidence that we wanted to take this program forward. And so that's going to be reading out Q4 later this year.
Okay. Right. So the phase 2/3 initiated 2Q of 2022. Safety preliminary efficacy in 4Q, I think, is what we're expecting. So any framing similar to the CCM, any framing you can provide here for, for expectations to read out?
Sure. Sure. Well, there's a number of endpoints that we're measuring here. There is progression-free survival. There is duration of response. There is time to progression. And there's also overall response rate. And like CCM, we believe that the movement in one or more of these measures will give us and KOLs confidence that, again, a next step is warranted for this program. And I think like CCM, you know, we'll be looking at the totality of safety data and efficacy data that we see with this trial and having conversations with the FDA around path to advancing it forward.
Okay. You have a, you know, a whole slew of assets behind these as well in the pipeline, REC-4881 in clinical studies for FAP and cancer. You also have RBM39, which is Target Gamma and ovarian cancer and your new fibrosis asset that you mentioned, Target Epsilon acquired from Bayer. Not sure where you want to begin, but any high-level thoughts on these programs?
Well, I think, you know, you frame it well. And I think big picture is that we're going to be having seven readouts in the span of about 18 months. And I think right there, it highlights, I think, the value proposition around AI-enabled drug discovery is that can you bring volume? Can you bring it consistently? Can you find novel insights? And I think that's what that pipeline represents. And you see that we've talked about CCM and NF2, but also behind that, we have phase two readouts in FAP, as well as AXIN 1 APC mutant cancers. That being first half of next year, we have a phase two that we're going to initiate in a C. difficile infection this year. We have programs in RBM39 for HR-proficient cancers, as well as Target Epsilon.
That's working through IND-enabling studies on the path to an IND. If I look at the totality of the pipeline that we have, I see breadth, I see depth, I see maturity, all of that from a platform that has plasticity across therapeutic area.
I guess just at a high level, it's a little bit of a step change in thinking when you approach a company like Recursion—like your internal pipeline, you know, you could have 20 assets, right? And they don't all necessarily need to work for the company to work. That's a little bit different than how investors typically approach biotech companies where the lead asset is the one, you know, and if that fails, the company goes away. How would you recommend investors be thinking about, you know, these updates as well as future updates from the pipeline?
Yeah. Well, I mean, as an investor by background, I both empathize and sympathize with having to track Recursion because it does seem like every quarter, every two quarters, there is a substantial new development that one has to wrap one's head around. And I think that's exciting, frankly. I think that's, that's a company I want to be a part of. And, and, and if I was on the investing side, something I would want to be following. I think that, you know, we talked about some of the data updates that we have coming, you know, this year. But I also think that there's a lot to be excited for because, you know, I think this space is evolving. And I think that these ideas are becoming more increasingly adopted by more and more companies and companies founded on these, on this premise.
But I think at Recursion, you know, we're going to have those seven readouts in the span of about 18 months. And that highlights a level of diversification around a pipeline. And of course, behind that, we have many other programs at the discovery stage, at the research stage that, you know, are making their way through as well. But to, you know, to complement that, there is, the potential for, you know, new trial starts, INDs. There's potential for, you know, program options from our partners. There's a potential for M&A options from our partners. There is the potential for new partnerships to be formed. And I think all of that highlights an operating system that is designed to bring novel biological targets forward.
It is designed to design novel molecules, to identify biomarkers, to target and stratify patient populations, and to do how we kind of begun this conversation, that we want to be able to increasingly move drug discovery and development to be that of a search problem, to be able to get that right drug to the right patient at the right time. I think as these ideas become increasingly entrenched in this space, I, I see this just increasingly being the way forward.
Okay. I, you know, we've got 1 minute left. I just want to sneak in a question about applying AI to clinical trial development because I think this is maybe a growing focus for you guys. Any, you know, high-level thoughts there what you're doing?
Yeah. Absolutely. Well, I think with respect to, you know, clinical trial design and enrollment, you know, I think we might be, you know, we might look to speak more about this, you know, in the future. But I think that, you know, how you can target patients, how you can enroll patients, all of that. I think clinical trials are far more sequential than they are cyclical. Like, so it becomes what parts can be compressed. Okay. And I think there's ways that one can use biomarkers to do that. I think there's ways that one can use, how to enroll patients, how to target patients, how to target PIs.
I even look at, frankly, we recently hired Najat Khan away from J&J and her role that she had within bringing forward some of the vaccine work, how she had data-driven approaches there that really helped to kind of galvanize their own clinical efforts to get patients, you know, with the vaccine work. You know, she's joining as Chief R&D Officer as well as Chief Commercial Officer. I think there's great opportunities for innovation there as well.
Fantastic. Well, unfortunately, I think we're out of time now, so we'll have to leave it there. But Michael, thank you very much for the discussion and for participating in the conference.
Absolutely. Well, thank you for having me.