Good afternoon, everyone. Thank you so much for joining us. I'm Salveen Richter, a biotechnology analyst at Goldman Sachs, and really pleased to be joined by Ben Taylor, Chief Financial Officer of Recursion. Ben, just to start here, your platform is integrated with AI and machine learning, with the goal of designing first-in-class, best-in-class drugs. With the clinical proof of concept you've reported to date, in what ways has your platform achieved proof of concept, and where do optimization opportunities lie?
Sure. No, great question. And I think one of the nice things that we have versus a lot of spaces in AI is we actually produce something physical. And so, for example, the chemistries, you can take to a lab, you can test their properties, and you can see how they compare to other chemistries. Or you can look at biological finding, and you can repeat with an experiment to be able to test it. And so we've had a number of different points of validation throughout the course. On the chemistry side, we've had eight compounds licensed by partners and move forward in addition to our own compounds moving forward. Most recently, last December, we had our CDK7 molecule, where the data that we saw in the clinic was actually within a couple of % of our predicted models for where we thought it would go.
And that's obviously a critical component where you have to get the PK/PD just precisely right to balance that therapeutic window for that target. So those have been some examples. I think also you can look at our partnership business as a whole. I mean, we've brought in nearly $500 million from the different partnership activities. Roche paid us a $30 million milestone for a first-of-its-kind neurology map. And we've seen they felt that there was incredible novelty coming out of the targets that are there in a space that is desperately in need of new targets. So I think we start to see all of the pieces coming together where we have the third-party validation coming in on both the biology and the chemistry side. And now we've integrated those together.
We also have a ClinTech business that we don't talk a lot about, but it's really focused on how do we do better patient enrichment strategies. We do our clinical trial simulation. We think through inclusion/exclusion, and we think that'll actually drive a lot of efficiency, but that's what's currently playing out, and so hopefully we'll be more validated in the near future.
You've been active on the business development front, notably with the merger with Exscientia recently, if I'm pronouncing that correctly. How would you characterize your BD strategy and your appetite for deals at this point, recognizing there's all those technologies that you could be looking to bring in-house?
Yeah, absolutely. Well, and so I came over from the Exscientia side. And don't worry, that name will. The fact that it won't have to be pronounced over and over again is a favor to everyone. But the Exscientia side, I think, really showed how we could bring those two businesses together pretty seamlessly. And in fact, our technology stacks started working together almost from day one. Now, that was because you had really what we believe are the leaders in biology with the legacy Recursion and the leaders in chemistry on legacy Exscientia . And so there was a great fit. We honestly don't see a lot of spaces where we need to do that sort of transformational acquisition again. I think a really great example of how you can find other ways to fill areas is the Boltz-2 partnership that we announced on Friday.
So that's a partnership with MIT, where it was not just being able to virtually model protein folding, but also really looking at the binding affinity, which is something that has not been done before in that space. And historically, you would use an FEP to be able to model it out. The problem is that's incredibly expensive and time-consuming. And so not something that you can integrate in with AI technologies. You can't have a 100,000 generative chemical screen where you put each one of them through physics-based processing. You would black out New York, you know, that used so much energy. And so what we were able to show is, in partnership with MIT, in a very cost-effective way, being able to go 1,000 times faster and roughly 10,000 times cheaper. So that is really a step change in being able to change that market.
We didn't need to do an acquisition to do that. We can do it in partnership and still have all the benefits. So I think we're trying to be very creative in how we go about business development. There are still opportunities. We will still consider them, but I don't think they're necessary.
What will we see overall from the pipeline over the next 12 months?
Yeah, so a couple of really important pieces. We just had an FEP data release at DDW, where we had an oral presentation. It was a very small patient set of the efficacy portion of the data, only six patients, but the depth of response and the time period to reach that response hadn't been seen before, and really importantly, we actually saw a reversal of dysplasia in some of the patients, which, if we can not only reduce the size and number, but also the dysplasia, that's going to change a lot of the paradigm for that patient population, and so what we want to see is that play out in a larger patient set and potentially also be able to do intermittent dosing, which would allow then better chronic dosing over time, so that data is coming around the end of the year.
I mentioned earlier we have the CDK7 program. We are going into combination studies right now. So we've got additional data coming on the monotherapy dose escalation, where we had seen some unexpected and encouraging efficacy data, but we really expect the combination to be where you see the stronger efficacy signals. So that's starting up right now, data to come. And then two other programs, RBM39, we would expect initial clinical data next year. This is where there's actually a great success story talking about the technology coming from the Exscientia side because CDK12 is a great oncotarget, but really, really hard to target with an inhibitor because the pockets for CDK12 and 13 are almost identical. Biologically, very different, as the humor of biology plays out, but the pockets are almost identical.
And so if you try and go after CDK12, you're almost definitely going to get a toxic drug. What we were able to show with the Recursion phenotypic platform is actually you can get the same biological reaction from RBM39 degradation. And so we were the first to identify that. We put it out, we published. There have been other groups that have published also making that connection since then, which is fantastic. And that program started in clinical trials late last year and would be a great not only validation of the phenotypic platform, but also just hopefully a good oncology drug. The other thing that we're trying to integrate into all of our programs, and that one very specifically, is a biomarker enrichment strategy. So important to always remember that clinical trials fail because of statistics, not because of biology or chemistry or anything else.
And so we do believe in finding that right patient population to try and give us the best statistical odds. That also means we could see that efficacy signal early on in the clinical trials because you're getting the right patients.
Great. Walk us through just the balance sheet at this point and how you intend to kind of where you can fund these programs into or where you might need to look at partnership opportunities?
Absolutely. So balance sheet is always an area of focus. And when we came in, the two companies independently spent $606 million in 2024, so quite a hefty amount. Doing a lot of different things, but quite a hefty amount. With the announcement that we had a month ago on the pipeline and strategic changes, and then today on our infrastructure changes, what we've guided to is a run rate, burn rate of less than $390 million. So that's about a 35% decrease over that time period. But we're actually still advancing a number of clinical programs, still executing and moving forward our partnerships and building out the platform. What we're trying to do is be very disciplined about data-driven decision-making and finding efficiencies. We are a tech company. We should be efficient. We should more with less every single year. And so we're going to keep pushing on that.
But that's been a key focus of the company and will continue to be. Hopefully, we are able to bring it down even further while still accomplishing the same things.
The company, they had a first-generation pipeline, and then we saw it emerge into second-generation and beyond. Are any of the first-generation pipeline assets still prioritized at this point?
FEP on the legacy Recursion side was first-generation compound. And I'd say there is a first-generation, which was like CCM. That was almost not a part of the platform, though it was phenotypic discovery. And then there was a couple of programs that were using an RNAi version of the platform, and FEP was included in that. What we're using now is all CRISPR-based. We've moved mostly over to Brightfield. And so it's a very different platform, but almost more importantly, we've layered in a lot more of models and multimodal data and different pieces that really help it stand out. But we are absolutely hoping the FEP program proves out. And even if it's not, I'm kind of a chemistry elitist, even if it's not the beautiful chemistry we always like to see, it is a patient population with a lot of need.
If we have good clinical data, that could be a real drug.
Do you see? I mean, does it seem or is there a plan here to transition the portfolio to being more of a cancer-driven portfolio just in light of what's played out?
Yeah, well, it's interesting. I think the right way to think about it is really on an aggregated risk profile, so if you think about back in January, we started guiding investors and saying, "We are going to do a bottom-to-top review of everything in the company," and not just pipeline, but operations as well, but what we looked at was, if we're doing a pipeline program, what's the patient need? What's the commercial need? What's the regulatory pathway? Biology risk, chemistry risk, all of the other things that could layer in, and so if you start to do that and say, "Where do I have the least variables I have to get right to get a successful drug?" it starts guiding you down in certain areas.
Some of the rare disease drugs are more difficult to think of in that aggregated risk sense because you could have multiple layers of regulatory, biological, chemical, commercial risk. That naturally guided us towards more oncology because it is a very clear pathway, very clear markets, very clear needs. I think we also want to always have drugs where we could take it forward ourselves. Many oncology categories are more fit for that than broader categories. You won't see us doing a diabetes drug on our own, for example. We might in partnership, but not on our own. That is part of why it feels more oncology-focused. Another part is it helps us build the infrastructure there. If you think about running a clinical trial, you need clinical networks. You need the right patient advocacy groups.
You need to know what good looks like, and that requires investment in infrastructure, so we wanted to bring more focus there, but we still, Sanofi is mostly I&I in our partnership. Roche is mostly neuroscience in our partnership, so we do a lot of different areas. The platform doesn't actually care.
So with regard to your cancer programs, you have a PI3K alpha preclinical program here. How are you thinking of differentiation with this asset in the context of what's playing out with some others, including Relay that has used machine learning and, sorry, motion-based design in their construct?
Absolutely. So I think there's two aspects that go into it. One is the selectivity. So what we're seeing is, at least in our hands, the selectivity is about an order of magnitude or more than the Relay or Scorpion compounds. And where that'll show itself is things like hyperglycemia, where you still see a significant hyperglycemia signal in the Scorpion product, for example. And so we have done some head-to-head work and believe that we are in the same category of potency as those drugs, but we'll have that order of magnitude better selectivity, which will remove some of the side effect profiles. I think there's also a lot of downstream chemical properties that people don't think about that never come up in the conversation. Actually, we as an industry are very good at making potent compounds.
We're okay at making selective compounds, and we're terrible at making compounds that have a good ADMET profile, and so that's a big part of it is how do you not fail clinical trials statistically because of side effects or absorption or metabolism or all of those different aspects? How do you actually test the biological thesis, and so depending on the drugs, some of them have cleaner downstream profiles, some of them less, but that's another area of differentiation where we always want our profiles to be very well-balanced and very clean.
You reported early CDK7 data. What do you see as promising in the data, and what are the next steps, particularly as you look at potential combination strategies?
Yeah. Well, that is actually a perfect follow-up for the last question because it's interesting. One of the most important elements there's a couple of different important elements here, but if you think about CDK7, it's a fundamental biological mechanism. It's upstream of CDK4/6, for example, great $9 billion drug class, but CDK4/6 only hits on one aspect of the cell cycle. CDK7 actually interacts with two different parts of the cell cycle and inhibits transcription. The transcription inhibition is really important because a lot of cancers respond to cancer treatment by upregulation of transcription. It's the most common escape mechanism. And in fact, in more metastatic cases, it's close to 50% have highly upregulated transcription. So you want to be able to inhibit both of those to set back on proliferation. So it should be a really nice biological target.
But because it's this fundamental biological mechanism, you're also going to have a lot of on-target tox if you don't have the right PK/PD profile. And so what we designed CDK7 to do was literally get in, hit your IC90, and get out of the system as quickly as possible. So we designed it to have about a 68-hour half-life. It also has to be cleanly absorbed. And this was really interesting. Looking at the other compounds, you had two major failure points. One was covalent inhibitors. You're almost definitely going to have a lot of tox with that. But the other was a lot of them were basically had efflux issues. So not something that normally comes up in biotech, but what that's going to mean for a compound that is a cell cycle inhibitor, you're going to be recycled inside of the GI tract over and over again.
So you're going to get a lot of GI tox. Patients don't like GI tox. It's one of the most common reasons to leave a clinical trial. So that means you're already starting with a statistical inertia in your trial patients that are going to leave because of that. You're also going to get highly variable absorption because sometimes it's going to be absorbed into the system, sometimes it's not. And so you can't have a viable dosing regimen. The last part is, it's ironic, but your tumor microenvironment is built to pump toxins out. And so even if you get there, it's likely to be pumped out. And so we looked at something like that and said, "You're going to have a bad side effects profile.
You're likely going to fail those trials, and you're not going to stay where you need to be." And so we designed it and then saw it play out in the clinical trials where we were in a couple of % on the half-life, and our absorption profile looked exactly how we would want it to with a nice dose escalation on the toxicity profile, which you would expect. And then we had a really nice not only PD signal from some of the biomarkers, but we actually had an efficacy signal, which we didn't expect. So this was a trial where all of the patients were actively progressing when they came in. Median was fourth-line metastatic disease. And we had a durable partial response in an ovarian cancer patient. We had multiple patients go from actively progressing lung cancer to stable disease.
Again, this is primarily a cytostatic, not a cytotoxic. Those were really exciting results to see. Now we're starting up a combo trial, and you'd expect to see it be more response-driven.
And then you have the RBM39 program that was designed to mimic CDK12 inhibition here while avoiding CDK13-related tox. Can you frame how this target was selected and what we'll see in terms of first data?
Yeah. And this is actually a great example of how the Phenomics platform works. So basically, the heart of Phenomics is to take a step back and break away from what's really linguistic biology, right? Target ABC1 is related to XYZ2 by this. But that's just not how biology works. It's much more regulatory and interactive. And so the Phenomics platform takes a step back and says, "I'm going to perturb a cell, see how it reacts, and then compare it to other cells that I've perturbed in different ways." So it might be a CRISPR knockout, might be a chemical perturbation, whatever it is. And so what that means is you can actually look and say, "This biological reaction and this biological reaction are highly similar." And what that allowed us to do was say, "I want something that looks like a CDK12 inhibitor, CDK12 knockout.
What else is going to give me that?" And it turned out RBM39 was there. And because we can also do the chemical perturbations, we can basically then say, "What chemistry is going to get me a reversal of that condition or can cause that condition either way?" We can test it either way. And so from that, you can see in a couple of minutes, literally, once the data is in Exscientia, you can go from, "I want CDK12, what's most similar to that, RBM39." Okay, if I want to target RBM39 and have this reaction, what chemistry is a lead that has produced a similar biological reaction? And so that, in a couple of minutes, can get you started on a program that then ended up being the RBM39 program that is now in the clinic.
That's very cool.
Yeah.
Just big picture, can you remind us where the collaboration with NVIDIA stands and any other collaborations you'd like to highlight from a technology standpoint and maybe speak to the cadence of economics from these collaborations on the forward?
Yeah, of course. So NVIDIA is a great collaboration. Not only did they come in help us build our supercomputers, which are actually a really critical part of how we're able to move so quickly is that we can go on site and do our own model training. And normally, supercomputer access is very hard to get and very, very expensive. And so it actually can be very cost-effective to have your own supercomputer. So they've been a great partner in that aspect. They're an investor as well. And in fact, we've had great engagement from them. Jensen's come out to a couple of our investor events and presented because he really believes that this is an area where AI will have some of the most practical applications in the near term. So that's been a really productive partnership, continues to be.
They were also involved in the work that we did with MIT on Boltz-2 that I mentioned earlier, and I think that's another great success and shows how we can all continue working together. As far as other technology partners, we're always looking for different paths. Tempus is one of our better-known technology partners that gives us access to a massive store of oncology real-world data, and we can actually use that, so if we see a signal in our Phenomics data, we can actually look at the Tempus data and say, "Do we correlate it or not? Do we see that signal in real life and from what's out there?" and if you do, you have pretty high confidence that there's something that you should investigate there, and if you don't, you just need to figure out a way to validate one way or the other.
But you can see how those things build off of each other, in addition to using Tempus for, obviously, our clinical trial work and patient enrichment strategies. We have a similar partner to Tempus, Helix, that we use for broader indications as well. So we're always looking for good data out there. I think we'll continue to be scrappy and creative in how we think about bringing new technologies and data in. And what we're really focused on is how do you bring it all together? So it's interesting. The Boltz-2 model is obviously open source. And you could imagine a business model some people might have to charge a lot for it. But this is one where actually putting it out in open source is terrific because an entire community develops around it that benefits us all.
And where we can drive really differentiated value is by integrating it into what is a much broader stack. And this is why we have always invested, both companies before integration and today, in having that full process. And this is where it makes a difference. There's no single algorithm. There's no single technology. It's how do you bring it all together to get to a wildly better result? And so, yeah, I think that we will continue to invest in those. As far as the economics go, I think that the economics will be more driven by the pharmaceutical partners than the technology partners. But we are always open to different constructs on it.
What program within your portfolio do you think has the highest probability of success?
I'll tell you which one has the least aggregated risk. How about that?
Sure.
I think the clearest program is if you look at something like MALT1 because that's been something where, with the J&J and Schrödinger programs, it appears that there is a pretty clear biological rationale there, right? Like we're seeing effect. However, both of those programs have hyperbilirubinemia, which, if you combine that with liver toxicity, puts you in a pretty dangerous place. And so that's an area where, from what we can see, we're not inhibiting UGT1A1, which is what those compounds do, and that's what's triggering the side effect. And so that would be one where we do think we've got a nice, beautiful compound with good potency and selectivity. And in relatively few patients, you should start to see, "Did we take that UGT1A1 inhibition out?" And we went into a very different chemical space, so it's not close to their drugs anyways.
But I think that'll really open it up. The rationale being, if you're going after B-cell malignancies, you're most likely going to be dosing along with some of the drugs that are already existing in the space. Most of those drugs cause liver toxicity. And so that's where you want to take a step back and say, "What's my actual end goal here?" It's probably not going to be to have a monotherapy MALT1. So I want to think about how is this patient actually going to be impacted in the real world? Because it's not enough just to design something that's a MALT1 inhibitor. I need to design something that's actually going to benefit the patient downstream. And that's where you need the cleaner safety profile if you're going to be using it in combination.
Right. Well, maybe a last question for you here. Is there anything that you want to highlight from the pipeline or just the overall strategy here that we haven't discussed?
Yeah. I think it's funny. One of the, we have a little bit different mandate than most companies. And I think we've talked about this a little bit before. We have a clear mandate not to become a binary risk company, to continue trying to change the way that drugs are designed and developed. And so what we've tried to do is build a business model and keep that balance. So we do have an internal pipeline that each one of those products now, after we've done the prioritization, we feel really comfortable about if it performs the way that we designed it to. There's a good regulatory path, a good commercial market. But we also balance it with this partnership business where we could continue to bring in a lot of capital that helps fund the company and develop the platform, but also has a standout value on its own.
We look at those compounds, and when we do the math, it's about half the MPV if we do it in our partnerships with Sanofi or Roche versus doing it in-house. But we don't have to invest our own capital off-balance sheet to get that done. And so we're trying to always think about how can we create a more scaled, risk-balanced company to do that. So with the actions like we had today and that we had a month ago, this is us showing that we are disciplined and data-driven. We are ambitious. We do want to change the industry. We do want to advance a pipeline that will make a real difference. But we need to do it in a very disciplined way to make it through all of the interesting things that life throws at us.
Definitely. With that, thank you so much.
Thank you.