Recursion Pharmaceuticals, Inc.

Learning Skull. I'm Chris Gibson, co-founder and CEO, and I'm excited to take you through Recursion's 2024, 2025, and the time ahead. So with that, we'll jump into the slides. And I want to just set the stage, first of all, to share a little bit about the moment in time that we're in right now. Recursion is leading this field of tech bio. We're at the frontier of this exciting opportunity to decode biology, to change the way that drugs are discovered and developed. And I think what we're seeing in 2024 at Recursion are leading indicators of what this inevitable future may look like. And what we're gonna see moving into 2025 now is a cascade of proof points that are gonna make it more and more obvious to everyone about what the future of the biopharma industry looks like. Now, to talk a little bit about how we've gotten to this place in 2024, I want to dive into some of the year in review, so to speak. And I want to kick that off by talking about our clinical data readouts. Last year, 2024, was the first time that we got to talk about efficacy in our first two clinical programs, REC 617 and REC 994. And I'll start with REC617. This is a potential best-in-class CDK7 inhibitor for advanced solid tumors. And during this sort of dose escalation phase of the study, before we would have expected to see efficacy, and in fact, in monotherapy, which also is an exciting opportunity for us, we typically would have expected to see efficacy with combo therapy. We saw not only reasonable safety, but we saw early signals of efficacy. We had one patient who had a reduction in their tumor that was sustained for six months or more and a number of patients who had stable disease. So this is really, really exciting. And we're actually kicking off combo studies here in the very, very near term. Rec. 994 was another program, one of the ones that helped originate Recursion, an oral superoxide scavenger for the potential first in disease opportunity to treat symptomatic cerebral cavernous malformation. And we demonstrated robust chronic safety treating these patients for a year or more. But also in this signal finding study, a study designed to help us identify early signals in efficacy across many different potential endpoints to design a future trial. We saw very encouraging reduction in lesion size based on the subjective MRI measure and also encouraging trends in functional improvement in the modified Rankin score in this phase two study. So very exciting from our perspective to get to start to turn over these cards of the earliest programs from the Recursion OS. What's more, we launched multiple additional trials, including our REC1245 trial for RBM39 degradation in solid tumors. We were able to launch our familial adenomatous polyposis trial, REC4881, and then also our C. difficile trial with REC3964. Beyond that, we continue to advance the next generation of molecules towards the clinic or into the clinic with CTA and IND updates for multiple programs, including the IND clearance for LSD1 in small cell lung cancer, the CTA for MALT1 and B-cell malignancies. We were able to initiate IND studies in our IPF program, REC4209, and initiated IND enabling studies in hypophosphatasia with our REV102 program, which was part of a joint venture with with RallyBio. Even more happened on the pipeline slide. You can see some of the deliveries that we made here in green. You can see that we were delivering across this robust pipeline all throughout 2024 and early 2025. And we're excited to continue that as we move into the coming year. But it wasn't just our pipeline. It was also our platform that gave us so many of the early new indications that we're advancing. One of those I want to talk a little bit about today, because over the last couple of years, we've started talking mostly about our clinical programs and talking less about some of the early programs in the space. There's just not enough time to go into all the incredible science. But given the Lily Scorpion deal we saw a couple of months ago, I want to highlight one of those early programs. This program, REC7735, this is a mutant-selective PI3 kinase inhibitor. And we used the Recursion OS to identify and optimize this molecule. And it's 100-fold more selective for the mutant compared to the wild-type in our early preclinical models. It's probably 10x-fold more selective than other wild-type sparing inhibitors that we've been able to test it against. And most importantly, we've seen this limited hyperglycemia when we've tested this. And I'll show you a little bit more of the data here. You can see on the left in vivo in this CDX model, we've taken our molecule, which is in sort of late stage optimization here, REC7735, and you see at a variety of increasing doses, this tumor regression, which is on par with the stx compound here this is the scorpion compound that garnered so much attraction uh so this is this is really exciting to see but what i get even more excited about is if you look in naive wild type mice at hyperglycemia this is one of the major challenges of the pa3 kinase inhibition space is that you can see compared to sdx in in these naive mice that were treated for five days with the molecules that the plasma insulin levels are dramatically lower in this REC7735 molecule, even at a dose of 150. That's twice as high as the highest dose you see over on the efficacy side. We think this is really, really exciting. We've had a lot of interest from potential partners as we first started talking about this in JPMorgan a couple of weeks ago, and we continue to optimize this molecule and advance it. This is just one of the exciting early discovery or advanced discovery programs that our teams are working on here at Recursion. Beyond the pipeline, the exciting programs that we're advancing, we're also in these incredible partnerships working with Roche Genentech, Sanofi, Bayer, and Merck KGA. I want to talk a little bit about the Roche and Sanofi partnerships. Those are the ones we're spending the largest amount of our time on. On the Roche side, we generated multiple whole genome phenomaps in oncology and neuroscience last year. That led to a $30 million milestone that was received. And I would say just a ton of excitement at the Russian NTECH side and on the recursion side as we start to look at novel, exciting biology in both oncology and neuroscience. On the Santa Fe side, in 2024, we advanced two programs through initial milestones. Those generated $15 million in aggregate payments. There's a number of additional programs that are advancing towards or through milestones in the near future as well. So we're really excited about our work there and continue to do really exciting work with Bayer and Merck at KGA. Finally, I want to talk a little bit about the work we did on our platform in 2024. We continue to lead the industry across data, across foundation models, across compute. And I think, really, it feels like it was much further ago than just last year. But we actually built and launched Biohive 2 last year with Nvidia. We believe this is the most powerful supercomputer wholly owned and operated by any single biopharma company. And I know as I watched the dashboard of utilization over the last year, we have been running this thing hard. We've been leveraging it to build a variety of different foundation models, to advance these models, and to explore new architectures of our neural nets across a wide variety of questions as we build sort of towards these world models of biology. Our data capabilities continue to grow and expand as well. We were able to map more than 1.4 million active ligands to binding pockets for structure-based drug discovery and target deconvolution last year. We've generated now up to 6.2 million multi-time point brightfield images each week on our Phenomics platform. And we produced just under a million transcriptomes last year, putting us at a total of more than 1.6 million since we launched this work in 2023. So really exciting to see these new data capabilities, these multimodal data capabilities advancing at Recursion. And of course, what we've been building in the discovery and translational space for many years, 2024 saw the beginning of our work in causal AI models and in ClinTech. So we used AI models on the Tempest data and Helix data to start building out causal understanding of biology from patient data for the first time. I think one of the most exciting projects here was deploying the Tempest data to build a patient stratification framework in small cell lung cancer. for one of our programs that we're advancing into the clinic there. And we've also started building automated site engagement tools and enrollment tools to accelerate patient matching, to accelerate patient recruitment, and essentially do a lot of the ClinOps work as we start to scale this pipeline at Recursion. So early days there on the ClinTech side, but really, really exciting as we build this full stack platform. So we've talked about our platform before, as companies move into the tech bio space, it's really important for us to be able to generate and aggregate data across many different levels of biology and chemistry. And I think recursion really leading the field in generating this real world data or aggregating this real world data. You can take those data, you can learn from them, you can build models, foundation models across all of these different kinds of questions you might ask in biology and leverage those to make hypotheses that you can then go back to the laboratory and test. And it's this iterative loop that makes us so excited that we think is going to be essential for advancing the field. And what we're seeing is that just looking backwards here, looking at recursion and Accenture alone, we've already been able to demonstrate these leading indicators of being able to quickly validate hypotheses compared to the industry average. So using this incredible set of real-world data and these world models, we're able to take an early hypothesis, validate it, and move to the hit stage faster than the rest of the industry. Using looking back at the legacy Accenture tools, we're able to take those compounds, those early hits and design them up to a more optimized stage, synthesizing dramatically fewer molecules than the industry average. And what that means is that we're spending less time compared to the industry average and going faster or we're spending less money and going faster compared to the industry averages. And of course, the most important measure that we want to change is the probability of success of our molecules in the clinic. And we look forward as we start to read out trial after trial in the coming years to being able to benchmark ourselves there and hopefully start to show some movement across the industry average probability of success. But I think these leading indicators are really, really exciting. And now I want to invite up Lena Nielsen, our SVP, head of platform, to join me here in Salt Lake City to talk a little bit about how now, not just looking back, taking our Accenture combination and looking forward and over the last few months since we brought the companies together. Lena, maybe you can talk about how we're seeing these benchmarks start to improve.

Yeah, thanks, Chris. So already as two independent companies, as Chris mentioned, we built two technology platforms that are already realizing great efficiencies in drug discovery and development from two complementary angles. At Recursion, we were founded around novel insights from complex biology mapping, so essentially building giant data sets fit for purpose for generating industry-leading foundation models of human health and disease. And then on the Excientia side, technologies, models for compound optimization, so efficiently synthesizing molecules with the right multi-parameter properties for potential treatments. And now we're bringing these two complementary sites together. And very specifically, at the merger, we created 90-day plans to very rapidly supercharge generating this real value. And we promise to tell you about the outcomes from that. And here I am with some of the first results, and there will be more to come later. Specifically, let's see if we can. Apologies. Specifically, as you can see here on the right, we have brought together massive data sets from both companies into a unified platform from admit, liabilities, phenomics, and cellular function, protein ligand binding, put all of this into our Centaur model management platform together with recursive massive compute that you heard about in order to generate a new generation of models that are more powerful, more accurate, and more generalizable. and then immediately deployed this using Centaur against our pipeline and the recursion OS. And within these short three months, able to quantify some of the benefits to the pipeline. So for example, deploying new scientific agents that have resulted in a 60% reduction in the human time needed to get to hit to lead initiation. Or a federation of models based on a new data set of over a million compounds for more accurate MOA deconvolution. Another set of models and centers also new with a two and a half fold increase in the efficiency of detecting new bioactive compounds. That's whole groups of new potential useful drug starting points and 40% reduction in likely cytotoxic compounds. That's just one example there among 18 new admit applications. So really already now seeing massive progress of building a new joint platform where the new hole is much greater than the sum of the prior parts. And with that, I'm going to hand it off to Ben Taylor, our CFO, who's going to talk about how we are doing all this new exciting work while at the same time, saving money as a joint entity.

Thanks, Lena. Yeah, it's been a great year for us, both coming together as companies as well as what we've seen looking forward. So for starters, we had 83 million in revenue as a combined group that's a pro forma basis in 2024. and had an ending cash balance of over $600 million. Now, that gives us enough of a runway to be able to extend into 2027. So continuing on the business model that we have, we feel very comfortable that we've got the runway to be able to execute on a lot of those things that Chris and Lena were just talking about. An important element of that, when we closed the deal, we gave guidance that we expected up to 100 million in synergies. We actually believe we will achieve a majority of those synergies this year and be able to get to a run rate that is beyond that 100 million over time. Now that is coming both from the more traditional synergies that you would expect from two companies coming together, but also on a lot of the operational synergies. So all of those benefits that Lena was just talking about actually translate to economic benefits for us too as we push forward. In addition, really exciting, important note, we have carved out the Vienna operations into a new company that will continue on and that both gives the new company a great standalone mandate as well as helps to organize some of our operations and really provide us as much focus as we can possibly get. And we're on track to also clean up a lot of the excess office and legacy sites that we had previously when we came together as a combined company. So we will give you a much more in-depth understanding of some of these developments in May. What we are doing right now is basically going through all of our operations and going from the bottoms up as well as doing a strategic assessment. And we want to give that to you all at once in a few months. Now, as we look forward into the next year, We are really, really excited because not only was 2024, as Chris would like to say, our best year yet, but when we look forward to 25, we just see so much coming through the pipeline that really starts to show signal and point that the technologies we're developing are having a material difference. Chris brought up CDK7 and CCM and the initial data there. We have a number of additional studies that are getting started and will have readouts in the near future that could continue to support that moving forward. I think the PI3K is actually a great example as well, because that is an example of how we have this deep pipeline that most people don't even know about. We were able to look at a situation that was in the market and move quickly to provide what we believe is a better quality candidate that we can move forward with. And it is in that area that is not normally viewed by the rest of the investor market. So really exciting to be able to talk about more of those pipeline programs coming through in the future as well. Partnerships will continue to be an important part of our business model. I really want to highlight that in a part of our revenues, We brought in 45 million in cash payments for achieving milestones from Sanofi and Roche. So that's very different from an upfront payment or a license to our technology. That is actually achieving technical goals that were extremely difficult and highly valued by those partners. And so we expect to continue achieving those milestones both generating cash flows, but also validating that the technology is doing what we want it to do. And then we will also continue to bring forward more of the data on how the platform is exceeding benchmarks, moving faster, being very efficient in how it does it. And so you can expect that through the course of 2025. All of that, I think, comes together. We take it very seriously. Our mandate from our investors is not to just be a biotech company. It's actually when we say tech bio, what that means is we want to change the underlying probability of success across the biopharma industry. And that's by creating better quality medicines and doing it in a more efficient way. And so we look at all of these points and try and stay focused on the bigger vision. And I'm going to turn it back over to Chris because he can really talk about how these points come together and where we see it going in the future.

Thanks, Ben. Yeah, I think we're so excited about 2025, as Ben just shared, so many different potential catalysts upcoming across our pipeline, our partnerships, and our platform. But as we look out a little bit beyond that, sort of to the intermediate or even longer term, we believe that there is just extraordinary value to be unlocked. And we think Recursion is better positioned than almost anyone to be able to do this work. And so I want to share a little bit of that vision for you. I want to talk a little bit about how we think about the intermediate term. And of course, recursion today is leading the field in bringing together the real world and building these world models. This is the idea that we can have a laboratory full of scientists who are generating data, that data can be leveraged in a computational environment to turn it into models. We can learn and hypothesize using those models and go back into the real world. And this has been what we've been building at Recursion for quite some time, this idea of the real world and the world model, this loop of learning and then hypothesizing. But what we think is coming, what we believe is coming in the near to intermediate future, is actually a transposition of these, a transposition where the world models become so good that they actually start to look more like a virtual cell. They start to actually look like a virtual cell, and that virtual cell is well-positioned to help us make predictions about biology. And so if we are taking this virtual cell and making predictions about biology, instead of taking real world data to inform algorithms to go back and make predictions, there's this transposition that we think is going to fundamentally shift the way that the field engages. It's going to allow us to explore biology more broadly, chemistry more broadly. And we believe that there's a number of different data layers and capabilities that are going to be required to truly build a highly, highly realistic virtual cell where you can simulate biology and chemistry. And at the very macro side of that will be patient models. These are real world patient data and AI. And recursion hasn't generated a lot of our own data in that space, although our clinical trials are one opportunity for us to do that, but we've partnered with incredible groups like Tempus and Helix, and they're allowing us to bring in tens of petabytes of data from patients all around the country and beyond, and to leverage those to build causal AI models with our underlying data. When we start thinking about the pathway levels of biology, Recursion, I think, more than any other company in the space, this is what we're known for. Hundreds of millions of different perturbations of biology in our laboratories where we've generated this fit for purpose data set across multiple layers of omics, We think we're gonna continue to lead there. Of course, the protein model side with AlphaFold and other protein folding models, we think this is so exciting to see, and we think it's gonna be relatively commoditized. There's so many groups on the cutting edge here, and we haven't yet announced all of these, but we are working with partners in this space to make sure that Recursion has access to some of the most advanced protein folding models in the world. And one step below that on the true micro side of things is the atomistic models. This is sort of the QMMD side of things. And some have said that AI will never play a role. This is AI can't really do physics. And I think we're going to see that that is dramatically incorrect. What we're seeing at Recursion is that AI brought together with QMMD is actually putting us in a position with the legacy Accenture team and all of our compute to potentially lead here. And there's a lot of companies and foundations and institutes that are working on this kind of atomistic side. But I think recursion is going to have some really exciting things to share in the coming quarters. But it's when you put these all together, it's this vertical stack of data across atoms, across proteins, across pathways, and across patients, where I think recursion is truly going to create this competitive advantage, integrating all of those data into a virtual cell. And while I don't think that's something we're going to be able to deliver in the next 12 months, I do think it's something that is not in the too distant future. And we're going to keep working very hard to win that particular race. So now that we've taken a look at 2024, 2025, and maybe a little bit beyond, I want to go and actually dive into questions. And we've got lots of great questions coming in. Going to turn here to our Q&A monitor. Looks like the first question is coming from Alex Dranahan from B of A. Recursion has spoken about its supercomputer and data scale, especially on the phenotypic side in the past. But new advances, including DeepSeq, are bringing the need for scale into question. I would question that. Do you think this is a risk for tech bio as well? Or is biology just so complex that scale will continue to be essential? Well, look, Alec, I think it's a great question. I mean, biology is extraordinarily complex, and the interaction of biology and chemistry is extraordinarily complex. So I think scale is going to continue to matter. And while DeepSeq did demonstrate that there are ways that you can train and deploy models in a more efficient way, uh it does not hurt to be able to bring scale to these uh bring scale to the latest generations of neural nets and architectures I think we're going to be able to do that and we're going to be able to bring data across all these different data layers together and so we don't think that there's a dramatic opportunity for somebody to essentially bootstrap biology and chemistry we just think that it's it is fundamentally too complex Next, we'll go to Vikram, one of our analysts from Morgan Stanley. How is your partnership with NVIDIA progressing and what are your key focus areas for your foundation model work? Great question, Vikram. So we've been working with the NVIDIA team for many years. We're working on a number of different projects and they've been helping us to deploy lots of our different models across these very complex supercomputers, Biohive 2 being the one that we have in-house at Recursion. It is not trivial to be able to train on a supercomputer of that scale, not just one or two different models, but many different models across multiple teams and multiple sites. And what's more, we're not just using the GPU side of things, we're also using the CPU side of that supercomputer to actually do some of the atomistic work that I talked about just a moment ago. So there's a lot of complexity in just deploying all of these tools, and NVIDIA helps us there. There's lots of work that we're excited to continue doing with the NVIDIA team that we haven't talked about yet, and we'll have to wait in future quarters or years to be able to share more of that with you. Now moving over to Gil at Needham. Given costs on compute appear to be going down, how much of a moat does owning your own supercomputer still offer? Yeah, this is a great question. Look, I think there's two things that are necessary and neither is sufficient to build these maps of biology and figure out how to advance medicines more quickly. One is data, and one is compute. And while the cost of compute is going down, it's not trivial to do compute at the scale that recursion is doing it. So over two, five, 10 years, you may see a dramatic reduction. But I think we're going to be moving to virtual cells before we see a 10X reduction in the cost of compute. So we think it will be important for companies operating at the kind of frontier of tech bio to be able to have access to world-class compute. And at least over the next two or three years, we think that's gonna be a competitive moat for recursion. And at the same time, the data side we think is the extreme advantage for us. Because it doesn't matter how much money you have or how advanced we get on the biology side of things, it still takes time for cells to grow. It still takes time for a CRISPR knockout to mature and create all of the effects downstream in a cell. And biology is so complex that there's sort of this binomial tree of potential possibilities that would take an infinite amount of time to test. And so this virtuous cycle of learning and iteration where you can test at some scale, make hypotheses, and go back and validate or improve those hypotheses, we think that's going to be key. And we think recursion is years ahead from almost anyone else in the space in terms of building these data or aggregating these data across all these different levels of biology. So we feel really, really good about that. Next, we've got some financial questions. So Ben, I'll turn these over to you. Melissa asks, can you go into more detail about the Q4 revenue drop in 2024 compared to 2023?

yeah this is a great question and and it causes a lot of confusion out there because we are in many ways a tech company however our revenue and earnings don't show up like a traditional tech company in the sense of we are often in our partnerships paid a upfront payment that then is recognized as revenue over a longer period of time so you can't think of this like a subscription or a payment that's coming in quarterly and so Important fact, we've brought in $450 million from our partnerships. A significant amount of that still has not been recognized as revenue. So as we continue to go on and have milestones or enter into new partnerships, those will continue to be cash inflows. They may or may not show up immediately as revenue. So it's really, really hard to track our quarter-to-quarter performance based on our revenue, and we never suggest that people guide to it.

The wonderful world of GAAP accounting. Thank you. All right. Let's go to Dennis Ding from Jefferies, who asks, talk about your expected cash burn for this year and what we should expect at the May 2025 update.

Sure. We've got another complicated accounting question. So because of when we closed the transaction, most of the financials in our 10K actually reflect the recursion the legacy recursion standalone financials with a stub period for the legacy accentia piece so what we tried to do was actually put in you'll see in the in the press release in the 10k the cash burn amount of 184 million That was the starting year and end of year cash balance for Accentia, the difference, as well as if you look at the cash flow from operations and the capex from recursion, you're going to get to a number that's a little over 550 million. uh for the combined entities that is not a perfect number by any means to be clear but to give people a general sense i think uh what we're comfortable with is we are going to continue to grow but we will be able to manage uh our cash burn and be underneath those numbers this year and we'll give you more detail in may i don't want to front run ourselves we want to run a proper budgeting process but we are very very focused on cash burn we are very very focused on runway and we'll come back to you with more guidance on that later as well as what makes up it and why awesome thanks ben

Next, it looks like we've got a couple of questions here on our CCM program. This is REC 994 and CCM. So Joe Phillips asks, any updates on timing on REC 994? I was wondering if there's more clarification on whether this is going to advance to commercialization. And then Jeff at BioVantage asks, the primary endpoint of safety for REC 994 was met, but it was negative with regard to efficacy. Can you comment on the status of that program and plans moving forward beyond the recent presentation? What is the rationale for the longer-term treatment that will lead to statistically significant improvements. So great questions here. So I want to take Jeff's question first, which is we did see really, really exciting, robust safety across this chronic treatment of a year. But I would challenge this idea that we did not see efficacy in the study. We did a signal finding study. We're the first company to ever go to any regulatory agency with a ccm clinical trial to look at efficacy and so we had to look at a wide variety of different measures a wide variety of secondary endpoints that could give us an idea of where to go in a subsequent trial and so in a signal finding study you don't necessarily power all of those different endpoints you go looking for maybe nearly significant or somewhat significant, but not p-value less than 0.05 findings that you can then parlay into a subsequent study where you narrow down the number of endpoints that you go after. And what we saw was, I think, nearly significant data with a poorly powered study across this objective measure of MRI. If you look in the brainstem lesions, for example, we see really robust reduction in these particular lesions. And you have to be careful looking at so many different measures. But we've got this long-term extension study that we'll be able to look at soon that'll give us insights into whether these trends are continuing. And then we also saw this trend in modified Rankin score, which is really, really important because there's a precedence at the FDA from a functional standpoint of looking at neurologic diseases and showing a reduction in this MRS score over time. And so these will be key endpoints that we hone in on as we have discussions with the regulators about how to advance this program. And you can imagine if we go forward with a smaller number of endpoints and a larger number of patients, we may be very well positioned if we saw the same quantity and quality of reductions in lesions and improvements in symptoms to actually show statistical significance now in terms of the the details uh later this year we will be able to come back to you post interaction with the fda uh and and post maybe some exploration of our long-term extension data with some more clear plans on how we plan to advance this program uh but i know today the signals we saw in in uh this signal finding study were ones that we're excited about uh and certainly worth exploring for a disease with no other uh treatment besides surgery All right, and we've got S Jane on the business strategy. So the company's revenue stream suggests a split between a long term royalty base of revenue and then viable drug candidates and short term partnership based deals with existing pharma companies. So there's a two part question. While we've seen a lot of gains in the partnership revenue, are there any plans to expand and diversify revenue sources beyond the current avenues? So maybe, Ben, actually, I'll send that one over to you and then I'll take the second part in just a minute here.

Sure. Well, and it's really interesting. We've certainly done the work internally to look at different revenue streams to come in. What I would say is the economics that we get out of our pharma partnerships are excellent, for example. We're looking at over 300 million in milestone payments per program, high single, low double digit royalties, and effectively we have all of our direct costs paid for upfront. And so that's a very economically attractive deal. For us to diversify into other areas, we actually have to exceed that sort of an ROI threshold. So we do continue to look at it. We do engage with, whether it's partners on the tech side or on the pharma side, different ideas. But I'd say there's a very high bar for us to expand that.

Thanks, Ben. The second question from S. Jane is how confident is the leadership in our ability to discover and successfully clear clinical trials and get viable drugs on the shelves, potentially tapping into some of the delayed longer-term royalty-based revenue. And I would just say, look, I think we're very confident here. Discovering and developing medicines is hard, but we've been building a learning system. And my strong belief is that the system we have built has a high probability of being able to generate better molecules and better medicines over time. And so whatever level we're at today, on average, I would expect each generation of new molecules to get better. Now, it's important to note that for any individual program, There are hundreds of ways that it can fail. And some of those can be really, really surprising. And so we can't say with any confidence whether molecule A, B, or C is more or less likely to advance. There's certainly ones where we're investing more to go faster. But it's very hard on an individual asset level to be extraordinarily confident when you're in sort of the preclinical stage, the phase one stage, the phase two stage. but what i can say is both we with our own pipeline and our partners with a number of programs that have been moving forward in those partnerships see a lot of potential and what i think is really important about recursion is that we're not a typical biopharma company that has a handful of assets that create almost a bimodal outcome for the company where if you're successful in that leading program in that phase two trial the company's bought for some really really exciting number and that molecule advances in someone else's hands We're building a platform that's going to allow us to take many programs forward through our pipeline, many programs forward through our partnerships, and that starts to remove that kind of bimodal risk. And if one believes our platform is at least as good as the industry average in discovering and developing medicines, but we're doing it at a higher scale and more efficiently, I think recursion can become a really, really important company in the space. If we can do that and over time start to demonstrate that we're increasing the probability of success. And again, we're not here for trying to build this company and sell it in the next year or two. We're here a decade in with decades more to go. But I think if we can start to demonstrate that improvement in the probability of success while generating molecules at scale and doing it more efficiently, I think we're a company that has the potential to truly transform this industry. And so those royalty-based revenues would be coming in in that case. And of course, so would revenues from our own programs that we develop with our internal pipeline. Awesome. Next, let's go to Gil again from Needham. Can you provide any visibility into milestone payments in 2025? Maybe I'll turn this one over to Ben as well. And maybe I'll add a little something, but Ben, take this one.

Sure. The short answer is no, we're not giving any guidance on it right now. But I think what you've seen come out of our partnerships in the past is milestones around either drug programs advancing or delivery of phenomaps. I would say we are doing both of those things still in our partnerships. And so we look to do more of the same and more of it. The other thing that I would note is, and this is alluded to in what Chris was saying, another thing that we are always looking at is when is the right time to either advance a program on our own or to potentially look at partnerships around them as they advance through. And so I think that's something that we continue to look into. for our pipeline and that could generate revenues and milestones as well.

Thanks, Ben. Just a couple of things that I would add real quick right there is, you know, we've been doing the work of these partnerships for many years and essentially priming the pump on the milestones. And so while we're not giving any formal guidance for this year, I think there's a lot of confidence in where we're going. And as Ben was just alluding to, we're also starting to, I think, get a lot of interest in some of our individual assets that are in our clinical programs. or preclinical pipeline. And so hopefully we'll start to be able to demonstrate again, different ways that recursion is generating revenue, generating credentialization of the platform and subsidization of the future pipeline that we're gonna build. Next, so this one from Marcel. I'm actually gonna turn over this first question to Lena. So a while ago, and by a while ago, Marcel, this is a few weeks ago at the JP Morgan Conference, time flies. We mentioned the goal of creating virtual cells to enable us to discover and develop medicines at scale, including potentially to be able to simulate clinical trials. Given the known issue of hallucinations in LLMs, are there concerns that virtual trials could also be prone to inaccuracies? particularly regarding cell drug interactions, and in the rare disease space where data might be limited, how would you validate the results generated from these virtual trials to ensure their reliability and accuracy?

Yeah, sure. Great question. We rely, of course, on LLMs in addition to lots of other different model architectures, transformers, G-flow nets. MoE is an architecture unique to recursion that you might have read about if you're more deeply in the machine learning field. And these come both with great power, different powers and be able to get accurate predictions and also sometimes liabilities and limitations and hallucinations and false positives and so on. So we build big benchmarking data sets to ensure that our models are performing to the highest capability that we are able to do in our hands with those benchmarking models. In addition, I think a really great important component here is that we have large, laboratory setups in vivo setups etc where we can validate insights that come out of these models so that the play that we're doing harking back to chris's points it's not that we're never validating anything uh in uh real assays and real experiments is that we're able to focus down our experimental assays to the most promising compounds the most promising insights So we can be incredibly comprehensive exactly where it matters most and not spend time and money on compounds that never were going to go anywhere and insights that weren't going to be the right ones. And so we're still doing this incredible important validation along the way before a drug gets put into patients. And then specifically around rare diseases, this is definitely, of course, a challenge, not just for recursion, but for the industry as a whole. And one component to this is building models that are not just focused at predicting that specific disease, but models that can understand broad biology so that we can validate and have model performance that are not just about one specific gene, for example, but about that gene in context with everything else going on in the cell and in that patient. And that is one way to bring confidence in local areas that are quote-unquote rare, but in context of all the complexity of human health and disease, including through data at the atomistic and protein level, through our massive data sets and phenomics, linking all the way into the Tempix, Helix, and other patient data that Chris mentioned.

So adding onto that a little bit, thank you, Lena. Gil from Needham asks, as it relates to creating a virtual cell environment, what would convince you that, or us, that we've reached something productive? And so this is a great question. And the reality is, I don't think this is gonna be us flipping a switch and all of a sudden going from not productive to productive. It's gonna be a gradient that we achieved over the next handful of years. And Gil, we're really focused on benchmarking. Our team at Valence is helping to lead Polaris which is a benchmarking initiative across the industry. We've done a lot of our models against the therapeutic data commons, predictive ADMET benchmarks. And what I think we'll be looking for is moving from models that give us insights into predictive ADMET or mechanism of action deconvolution or clinical trial simulation, towards broader models that have these emergent features that help us ask and answer questions across many different layers of biology. If I were to just sort of imagine a place in the future that would give me the sense that we were really on the cusp and starting to feel like we had the virtual cell, it would be when we start to report to all of you a reduction in the scale of data that we're generating. And we're moving from scaled data into really just validating predictions. And that will be the point at which that transposition has happened. So, you know, we're still doing up to 2.2 million phenomic experiments a week. We've done 1.6 million transcriptomic experiments over the last year and a half. We're building a number of other kind of data modalities. When you start to see those numbers go down purposefully because we're just validating simulated outputs, and we're continuing to build the pipeline, that's when you'll know that we have that virtual cell. Perfect. I think there was one more question here from Laura asking about NIH funding and whether changes in NIH funding might impact the direction that recursion is going or our ability to discover and develop medicines. And what I would say is in the short term, No, I mean, recursion in the early days was dependent on NIH funding, both, you know, in my graduate school work with my co-founder, Dean, that was R01 funded research that led to some of the ideas that helped us build recursion, but also small business innovative research grants that we were able to bring in about $3.5 million in the early years of building recursion. I do worry about other startups in the space who may not be able to access some of those funds and what that means for the environment around us. And so, as you may have seen, Laura, we announced that building off of the incubator that is called Altitude Labs that is funded by Recursion, A number of entrepreneurs, myself included, and other entrepreneurs in the space are anchoring a fund aimed at bringing really exciting companies that might be impacted by this disruption in NIH funding to Salt Lake City to build and grow in our incubator to perhaps help them bridge that gap. Speaking a little bit more broadly, though, I do think it's very concerning. We see a number of institutions that have started deferring graduate student admissions. And while some companies are doing PhD and postdocs within their walls, all of the companies in our industry are reliant on the incredible grad student and postdoc a group that comes out of academia. And so I am very concerned over 10 or 15 years if we don't remedy some of these funding cuts that the US could lose its really substantial lead. Remember, most of the world wants to come here to train, and there's a reason for that. We're going to start to see that shift if we don't remedy some of these cuts on the NIH side pretty quickly. So great with that, I think we're probably going to decide to move on here. Thank you everybody for joining us for learnings call. We're really excited to be building towards the future here, decoding biology to radically improve lives and hope you all have an amazing day. Thank you so much.