Good morning. Happy New Year, and welcome to the 43rd Annual J.P. Morgan Healthcare Conference. I'm Eric Joseph, Senior Biotech Analyst with the firm. And it's my pleasure to introduce one of our first presenting companies this morning, Recursion. And presenting on behalf of the company is CEO Chris Gibson. Before I give him the podium, I just want to make a note that there is a Q&A session after the presentation. If you want to ask a question, just raise your hand, we'll get a mic over to you. And for folks that are tuning in via the webcast, just click the button to submit a question via the portal. So with that, Chris, thanks for joining us.
Thank you so much, Eric. Thank you to J.P. Morgan for the invitation. I'm delighted to be here. We're kicking off nice and early on Monday morning, and today, I want to talk to you about how we are decoding biology to radically improve lives. That's the mission of Recursion. It's something we take extraordinarily seriously. Of course, I'm going to be making reference to the future, and so it's important you understand the caveats that are listed here. Before I dive into the pipeline, the partnerships, the data, and all the ways that we are creating value, I want to set the stage for you with how we think about biology, how we believe you will decode biology, and we think that what's necessary is to create a virtuous cycle of learning and iteration based on real-world data and models that learn to simulate that real world.
And so what does that really look like? It looks like this. These are pictures from inside Recursion. You see on the left us generating massive quantities of real data across hundreds of millions of perturbations, generating rich, high-dimensional omics data, phenomics, transcriptomics, proteomics, and now increasingly different kinds of chemical data, automated chemical synthesis, etc . And on the right, you see BioHive- 2, the fastest supercomputer in all of biopharma, owned and operated by Recursion, where we are building ML and AI models that are taking that data and learning to understand and identify patterns in those data so that we can increasingly predict all of the experiments we have not yet run. And I think what we have done more than any other company in the space is verticalize this approach.
So we began with a point solution of phenomics, and today we are generating or aggregating with our partners data from patients to cells, cells to organoids, organoids to animals, and animals to people in the clinic. We're recording all of that high-dimensional data and training many different kinds of machine learning and AI models, building foundation models across many of those layers. And we believe that this is the way that we can industrialize drug discovery. The early evidence of all of that work, you can see here in a summary of what sort of value we've brought today. We have 10 clinical and preclinical programs. I'll be able to talk about a few of those in a minute. We expect roughly 10 milestones from those programs over the next 18 months. Beyond that, we have 10 advanced discovery programs that are moving rapidly towards the clinic.
We are truly building a pipeline of pipelines across rare disease, oncology, and other areas of unmet need. All of those coming from one unified Recursion operating system, and we're not just building our own pipeline. We're building pipelines for our partners, working with incredible biopharma partners like Roche, Genentech, Sanofi, Bayer, and Merck KGaA, and to date, we've been able to bring in over $450 million through those collaborations, not just in upfronts, but in milestones that we've earned, well on our way to achieving up to $20 billion in milestones before royalties, and all the while learning how to do biology, how to do drug discovery and development even better through the incredible partnership with these companies, so let's dive in and talk a little bit about the platform and how it's powering our portfolio.
Here's an overview of what we're working on in the clinic, in oncology, in rare disease, and in other areas. I'm going to walk through a number of these programs where we expect data in the near term, where we're really excited about the program. But I also want to highlight, just given some of the exciting leaks and discussions over the weekend, in the advanced discovery program stage here, just sharing for the first time that we actually have a really, really exciting mutant-specific PI3 kinase inhibitor that I think has become quite the talk of the town, this particular area of biology over the weekend. And we'll be advancing this very, very quickly towards the clinic as well. But let's talk about specific programs beginning in oncology. I want to talk about our RBM39 program, REC-1245.
This is a highly selective RBM39 degrader for biomarker-enriched solid tumors and lymphoma, and what I love about this story, which I'll tell you in a minute, how we identified not just an exciting molecule, but an entirely new pathway to allow us to take on what otherwise would be considered CDK12 biology. The entire R&D campaign from identifying a biologic action point adjacent to CDK12 or upstream of CDK12 to IND-enabling studies took us 18 months, and I think we're going to be able to do this over and over and over again for many programs, so let me tell you this story of how we identified an RBM39 inhibitor in this context. As some of you may know, we build these giant maps of biology.
We knock out every gene in the genome, and we do transcriptomics or take microscopy images and turn those into phenomics representations of biology. You can cluster all of those different experiments by similarity or difference. We can also put millions of chemical compounds into these same maps. What you're seeing here is a subset of the genome. Across the diagonal is every gene compared to itself, and across each axis is every gene. If you look at that little tiny red dot that's been expanded out, that is where we find CDK12. CDK12 is interesting. People have been very interested in the DNA damage response pathway and its modulation by CDK12. The challenge with targeting CDK12 with a traditional small molecule is that it is structurally very, very similar to CDK13. It is almost impossible to have a specific CDK12 inhibitor.
Now, what's interesting is that CDK13 has a very different set of functions. And so if you hit CDK12 and CDK13, it leads to a whole bunch of additional off-target toxicity that you don't want to see. When we look at our maps of biology, we actually see this. You can see CDK12 in red and CDK13 in blue. It indicates that functionally, knocking out CDK12 and CDK13 do very, very different things. And so the question we asked is, is there any other way to impinge on the CDK12 biology indirectly without modulating the CDK13 biology? And what we identified just by looking at maps like this across the genome and millions of chemical compounds was a series, now REC-1245, that was clustering with CDK12 and RBM39.
And that was interesting to us because a couple of years ago, there was no published literature suggesting that RBM39 and CDK12 were associated. So this was an exciting hypothesis for us. Was RBM39 actually controlling CDK12 biology? And was REC-1245 and the associated chemical series potentially either a specific CDK12 inhibitor or an RBM39 inhibitor? And it turned out through biochemical assays, we were able to demonstrate that REC-1245 is indeed a robust RBM39 inhibitor and has no effect on CDK12. And so if our hypothesis holds that RBM39 is modulating CDK12, perhaps we could see something in studies that otherwise you would expect CDK12 to show you some really interesting findings in. So here's a CDX model. You can see tumor growth going up in vehicle.
As you increase the dose of REC-1245 from one to three to 10 mg per kg, you see this dramatic reduction in tumor growth. Now, that is not something that we would have expected if you weren't modulating CDK12. What's interesting is if you also measure RBM39 % degradation, you see this nice increase from one to three to 10. Through a wide variety of other experiments that we don't have time to go through here, we convinced ourselves that RBM39 is leading to disruption of the DNA damage response pathway through RNA splicing changes that are driven by degradation of RBM39. That we've identified a really exciting novel RBM39 inhibitor. We drove this program through preclinical studies, through IND-enabling studies. The first patient was dosed in a study just last quarter.
Recruitment is active and ongoing, and we look forward to giving an update next year on this particular program. This is a program we are very, very excited about. New biology, new chemistry, and hopefully a way to address really high unmet need in oncology. Let's go to a program that's a little bit further advanced. This is REC-617, also in oncology. This is a CDK7 inhibitor. And this particular program relied more on the model side of our operating system. This was really a precision chemistry approach. And I think what's most notable is that this team went after CDK7, a very, very challenging target for a variety of reasons. They decided to look at multiple different parameters of that chemical or that chemistry that they could change to make the medicine potentially have a much better therapeutic window, to be much better potentially for patients.
And they were able to go from initiation of that project to a candidate ID and only synthesize 136 molecules. So that is compared to the typical 1,000, 2,000, 3,000, or more molecules that are synthesized in most traditional biopharma campaigns when you're working on novel chemistry. So very, very exciting there. We're doing dose escalation studies now. This is in a really challenging setting of patients who have a median of four prior therapies that they have failed. And it's a very heterogeneous population as well. We're trying to find the very best place that we can take this particular molecule. We've dosed up to 20 milligrams once a day. We've not identified the maximum tolerated dose. We've initiated twice-a-day dosing. I'll talk more about why that's important in just a minute.
Why we like this molecule, why the team we're confident they achieved the design parameters that we set out to, is demonstrated here. CDK7 and CDK2 are classically linked, and it is very, very challenging to hit CDK7 without hitting CDK2, and what you can see here, this is CDK7 IC50 in gray, CDK7 IC80 in black, and then CDK2 IC80 in red. You can see at both of our doses, we are well over the CDK7 IC80 and well below the CDK2 IC80. That's actually really, really important, and we're not sure that other CDK7 inhibitors are going to have anywhere near this sort of profile.
What's more, if we look at some of the other CDK7 inhibitors and we show data from those against the IC80 coverage, you can see that our molecule, REC-617 at 20 milligrams once a day, is dramatically covering the CDK7 IC80, whereas other CDK7 inhibitors do not appear to be. What's more, this molecule was designed to have a relatively limited half-life, and that's really important because if you completely inhibit CDK7 for a very long time, or you have a covalent CDK7 inhibitor, we are very confident you will see a lot of toxicity, and so by having a short half-life, really robust inhibition of CDK7 without modulating CDK2 anywhere near what some of the other molecules may do, we believe we've put ourselves in a situation to potentially have a molecule that can go after this target with a much broader therapeutic window.
And we're playing that out right now in the clinic. What I think we're most excited about is some of the very early data, frankly surprising data from the dose escalation studies. We didn't expect to see monotherapy efficacy. We have this very heterogeneous set of patients. But we have seen in this particular patient, after four prior lines of therapy with metastatic ovarian cancer, we were able to identify a patient who has a partial response that was achieved at week 16. They now have an ongoing response six months into treatment. And they have reduction not only in the size of a couple of these lymph nodes, both para-aortic and mesenteric, but they also have this very robust reduction in these tumor biomarkers. Again, this is a patient who failed four prior lines of therapy.
And we're working with monotherapy here when our intended use is probably actually to go to combination therapy. So we were very excited to see this. As I mentioned, we have not hit the maximum tolerated dose, and we are already seeing partial responses. So we'll continue with the dose escalation, and then we'll be kicking off combination studies in the first half of this year. And the last thing I'll mention, we've been building ML and AI tools for drug discovery since 2013 when we began. We were one of the early old-school TechB io companies. Today, we have started building technology tools not just through discovery and translation and even preclinical. We're now building AI tools in clinical development. And we're not going to share a lot about those today, but it fits with our thesis of building the full stack of AI-enabled drug discovery.
We have leveraged for the first time some of these new tools and new data partnerships that we'll share in the coming months and quarters to enable us to do some really interesting patient stratification. We have a really good sense about which patients we're going to want to continue putting into this study. Let's take a shift and move from oncology into rare disease. I'm going to talk about REC-994, which is a superoxide scavenger in cerebral cavernous malformation. If you haven't heard of CCM, this is a disease that affects six times the number of patients in cystic fibrosis. There is no medical treatment. We are the very first company to take a program into an institutionally sponsored study with the FDA. We did a year of treatment.
The primary endpoint in this study was safety and tolerability because if we were eventually to take a drug like this to the market, we imagine patients may take this medicine for life. And we were really, really thrilled to see that the drug was very, very safe. There are essentially no differences between placebo and the highest dose of the drug across a year of treatment. And I think there's a strong indication of that when you see that the long-term extension has enrolled roughly 80% of the patients that were in the trial. They've stayed on the long-term extension. What's more, in the exploratory efficacy readouts that we were looking for, trying to find a place that we could work with the FDA to take as an efficacy readout in a following study, we identified both time and dose-dependent trends in lesion volume reduction and hemosiderin ring reduction.
When we talk to KOLs in the space, especially KOLs working with patients who have these lesions in the brainstem or other very, very elegant structures of the brain, they believe these are really, really meaningful. We think they will lead to functional changes in these patients over time. We look forward to presenting the full results of our initial study in this disease in just a few weeks. We have a late-breaking oral abstract at ISC on February 5th. So if you're going to be there, we look forward to sharing with you. Another exciting rare disease program is REC-4881. This is in familial adenomatous polyposis. This is a disease associated with disruption in the APC gene. These patients get hundreds or thousands of polyps in their gut.
They have to have their colon removed typically in their late teens to early 20s, or else they will, with 100% probability, progress to cancer. We identified this interesting interaction of a subset of MEK inhibitors and APC-driven polyps generated really compelling preclinical data, and we've brought this medicine into the clinic now, and what we can share today for the first time is that the initial four milligram QD dose is both pharmacologically active and well tolerated, and we believe that we're going to be able to modulate this disease because this particular MEK inhibitor has a differentiated profile from an ADME perspective that enhances exposure in the GI, and so we're working through the therapeutic window now, and we'll be able to share some phase 1b/2 safety and early efficacy data in the coming quarters. A program we're really, really excited about at Recursion.
Another program, a newer program that we haven't talked about too much in the past, this is REV-102. It's part of a joint venture with Rallybio. This is an ENPP1 inhibitor designed to go after hypophosphatasia. So some of you may know there's a drug called Strensiq. It's an enzyme replacement in hypophosphatasia. It is a very effective medicine for severely affected patients with that disease, but it is very expensive on the order of millions of dollars a year. And it's only available to a very, very small sliver of the patient population. In partnership with Rallybio, we have designed and built an ENPP1 inhibitor that is demonstrating in preclinical models robust rescue of a variety of different readouts that are on par with the enzyme replacements. But this is a relatively easy-to-produce, small molecule. And so we are really excited about the potential here.
And I will just say, this is a program that has garnered a tremendous amount of interest from pharma partners. And one of the beautiful things about building a pipeline and not being a single asset company is that Recursion is positioned to potentially identify programs such as this that we could put in the hands or jointly develop with other companies that allows us both to take some burn off of our plate, but also allows us to learn and grow and get incredible medicines to patients across a broader set of therapeutic areas than we might otherwise be able to do. So very, very excited about this. IND-enabling studies are kicking off imminently. Outside of rare and oncology, just want to talk about one program here. This is REC-3964. This is a C. diff toxin B inhibitor.
We were able to leverage our platform to go after a modality-agnostic approach to recurrent C. diff. We identified that inhibition of toxin B, actually targeting the toxin itself, had a really robust preclinical effect. We went head-to-head with bezlotoxumab and showed superior efficacy there. Excited, we went into the clinic, and we've begun dosing patients in the phase II. The phase I demonstrated really, really, really good safety profile.
There were essentially no treatment-related discontinuations or significant adverse events, which is what you'd want if you're going after a patient population like you might have in recurrent C. diff. Again, just a little bit of a hint here, we've used some of our new ML and AI tools and our new data partnerships to rapidly, in weeks, identify 30 new sites that we believe we can bring on that are unique from the sites that are typically used for C. diff.
And this is, again, just an early indicator of some of the work we're doing there that's going to help us accelerate the clinical operations to underlie the 10 programs and more that we have in our pipeline today and will in the future. So I want to talk a little bit about our partnerships as well because we're not only building our own pipeline, but helping to build the pipelines of our partners. And we are very, very fortunate to be working with four of the best companies in the field. Roche/Genentech focused on the whole of neuroscience and one GI oncology indication. Sanofi focused on both oncology and immunology. Bayer focused primarily on oncology. And then Merck KGaA focused on oncology and immunology. And we don't have time to go through all of the partnerships today, but I want to highlight our two flagship partnerships.
First, with Roche and Genentech, we're able to share for the first time a little bit more of the progress we've been making with our collaborators at Roche and Genentech. And I'll start with the GI oncology side. We're able to share today that we have now generated multiple whole genome phenomaps with chemical perturbations, so hundreds of thousands of chemical perturbations in multiple disease-relevant cellular contexts. What this means is that we can now look across the genome and hundreds of thousands of molecules within multiple contexts in GI oncology to identify similarities and differences in the way those different cell contexts respond to different gene knockouts or to different chemicals. We believe this is really, really exciting. I think our partners at Roche and Genentech agree. Everybody's sitting around just wanting to explore these maps at rapid speed.
We already have our first validated hit series now in hit to lead, and my expectation is that with all of the incredible substrate in this map, the frequency with which we can validate these findings, that we're going to have additional near-term program options in this space, so that's very exciting to us. On the neuroscience side, we've worked for three years now to build the world's first whole genome arrayed CRISPR knockout map in neural iPSC cells. That is a feat that, as far as we know, has not been achieved anywhere else. We have rich omics data across the entire genome and many other perturbations, so we can essentially cluster the genome just like you saw in the RBM39 slide, but in the context of neural iPSC cells. It is extraordinarily exciting to look at that map.
We are seeing relationships that we believe are very strong and very robust together with our partners at Roche and Genentech, who can also query that map, that are giving us just a lot of excitement about where we could go in some of the major areas of neuroscience where there have not been that many new targets for a very long time. We were paid $30 million for that map. There are potentially more maps to come, and we already have multiple target validation packages underway with our partners, so we are very, very excited about what this may mean for the future in the large, intractable area of neuroscience where I know all of us believe we need to find new and better medicines.
On the Sanofi side, we've already had three programs that have advanced through milestones just last year, $15 million in payments for two of those programs that we're advancing. The relationship with Sanofi is very, very close. We were able to share one of our internal programs with them. They got excited enough about it. It actually came into the collaboration. So we're seeing crosstalk in pipelines, which is pretty exciting. And this year and in the near term, we expect to identify new targets. This is something that I think Recursion can bring to this collaboration that was initially built with Exscientia. So bringing new targets into the inflammation immunology space, new options for programs, and advancing additional programs into lead optimization are all on the table.
And I think both the Roche and Genentech and Sanofi collaboration are going extraordinarily well, and the Bayer and Merck KGaA collaboration is going well also. Now, before I close with some business updates, I do just want to come back to the broader philosophy of how we see the world. I started with this virtuous cycle of learning and iteration where we have these massive automated wet labs and partners like Tempus and others who are helping us build and aggregate the largest biological and chemical perturbation dataset in the world. And we are using that as substrate to train many different kinds of machine learning models to help us understand things like how can we predict ADME properties, how can we predict protein-ligand interactions.
What's happening right now for the leaders of the field, this is not kind of the rest of the industry, but for the leaders of TechB io, is that we're right on the precipice of an important shift. As these models become really, really good, as they become highly predictive, you essentially can build what we'll call here a virtual cell. So the world model becomes a virtual cell. And when you can build a virtual cell, you have a shift. All of a sudden, you have this transposition where your wet lab becomes the validation tool for the simulated output of a virtual cell as opposed to what's been happening for the last 10 years for those on the forefront, which is your wet lab is a data generation engine to build one day a virtual cell.
Now, this may not resonate with many of you, but I believe very deeply this is what the future of our industry will be. I think there are very few companies on the cusp. The reason we think this is so, so hard to build is that to create a virtual cell where you can simulate biology in a robust way, you need data and algorithms at really three different layers. I'm going to start at the top. You need pathway-level data. This is a place where I think Recursion is far ahead of the rest of the field. We've knocked out every gene in the genome of dozens of different cell types. We've put millions of chemical perturbations on top of these cells. We have rich omics data from phenomics to transcriptomics. We have time course data increasingly with bright field imaging.
All of these datasets are giving us a very, very robust understanding of the gene networks that are at play in many different cellular contexts. We are essentially building a map of the pathways of biology in different cellular contexts. Now, the protein layer is super, super interesting. AlphaFold kicked off a revolution. What's different about this layer is that because the data that spawned that revolution was public, we see many open-source competitors, Boltz, we see Baker Lab and Rosetta, and some of these other protein folding tools that are coming out. We believe this layer will be commoditized, and that if you partner with the right group, which we are doing, you will have access to state-of-the-art frontier protein folding, protein-protein interaction, protein-ligand interaction models, and so we believe we can be at the forefront of that field through partnerships.
Now, the last layer that hasn't been talked about as much is this atomistic layer, really looking at sort of QM/MD simulations, understanding of the atomistic layer, what's happening, and before the partnership with Exscientia, we were excited to one day go into that space. We had the fastest supercomputer in all of biopharma, including a massive set of CPU compute, which is important in the QM/MD space, and an incredible AI team, but we didn't have the data substrate. Through our new relationship with Exscientia, we brought on their QM/MD team, their QM/MD datasets, and we can't share everything today, but in the coming months, I think we will be able to tell the world that we are the leaders in this layer as well.
Recursion, I believe, is uniquely positioned as the leader at the pathway level, soon to be the leader at the atomistic level, and through partners at the forefront in the protein level. By combining those three and building not just foundation models, but a complete model of a human cell, I believe we will start to see in the coming years this transposition where our virtual cell is allowing us to make our wet lab a validation machine for just a kind of unimagined number of hypotheses. We are really, really excited about that kind of transformation. I want to finish with business updates. As you know, we completed the business combination with Exscientia just a few weeks ago, I guess, at this point, a month and a half ago.
And we're very, very excited to bring on new members of the team, Dave Hallett, Ben Taylor, our new CFO. Please find a chance to meet with these folks. We think we've got just an extraordinary team at the executive level, a fantastic board bringing on Franziska. So we feel like we are well-powered at the board and executive team level to kind of venture into this new area of unknown. And we are deeply focused on maximizing the ROI of what to date has been the largest acquisition or business combination in the TechB io field. Just a few kind of updates here. We had a multi-day in-person event in London. We brought the entire Exscientia team together with about 50 different leaders across Recursion. We introduced the whole group to the senior leadership teams that have been combined from both companies.
We are really starting to operate as one company just a few months after the transaction closed. We're excited to also share in our year-end earnings updates from our 90-day goals. Every time we do an acquisition, we challenge the teams to deliver on some early 90-day goals that will help to demonstrate the utility of the dataset or the people that we're bringing together. So we'll be able to give you updates on that at year-end earnings. We are working through identifying and deploying the synergies that we have promised in order to make sure that we can continue with this robust cash position, which at the end of Q3, we had a pro forma cash position of $752 million. I'm going to end here by just talking about how excited we are for 2025. 2024 was great.
Lots of stuff happening, especially the business combination with Exscientia was huge. We had multiple efficacy readouts that we took to be very, very positive, and now looking forward to 2025, catalyst after catalyst in the clinical pipeline, exciting new programs we'll be able to talk more about, as I mentioned before, like this mutant-specific PI3 kinase inhibitor. On the partnership side, we expect new phenomap options, new project initiations, program options, and potentially new partnerships, and then on the platform side, as I've shared, we're going to continue sharing more of what we're building from an AI and clinical development perspective. We're going to share more about these foundation models at those different levels that I talked about with the community.
And we're going to be continuing to integrate technology with autonomous workflows to make sure that the incredible employees of Recursion are using their incredible minds on the things that they do best and that we're using autonomous discovery and agentic AI for all the things where they could be really, really good. So I'm looking forward to 2025. I hope you all are too. Thanks for getting up early. A packed room for 7:30 A.M. on Monday is much appreciated. And with that, take any questions.
Okay, great. Well, thanks, Chris. And again, for those who have questions, just raise your hand. We'll get a mic over to you. But maybe just to start off, Chris, just picking up from the business combination that you highlighted, if we're right now, we're kind of, as we look at your present pipeline, it's sort of a marriage of the two former separate entities. I guess, how do we think, how should we be thinking about sort of the synergies that the combination with Exscientia sort of bring to the table, and sort of where would they possibly be reflected in either your existing or growing expanding pipeline?
Yeah, thanks, Eric. It's a great question. So I'm going to overly simplify for speed here, but I think Exscientia had built an incredible chemistry platform that was allowing them to really generate best-in-class programs quickly across really challenging multi-parameter optimization situations. And Recursion had built sort of a biology-first platform that was allowing us to find first-in-class biology. What we hope to do if we bring the companies together in the best way is to find programs that are going to be simultaneously both first-in-class and best-in-class. Now, what that doesn't mean is if we see an opportunity where our platform can be deployed to generate, for example, a new best-in-class in an area we're excited about, even if not first-in-class, if we believe the Recursion platform can do that, we will still continue to do some best-in-class programs. But we really want to bring the two platforms together.
There are fantastic synergies. If you look at the CRO bill at Exscientia, it was mostly biology. And if you look at the CRO bill at Recursion, it was mostly chemistry because they built an automated chemistry platform. And so we hope, sorry to the CROs, but to dramatically reduce the amount that we're outsourcing to those kinds of partners. Those are the kinds of synergies that we expect to see. And I think having the biology and chemistry across that full stack, and frankly, a team who's taken programs from start to clinic from those two companies, that's a rare set of people in the TechB io field. And we've now brought just even hundreds more of those people together. So we're very, very excited.
Thank you. For your virtual cell model, which seems pretty exciting, you mentioned for the protein side that you were either partnering already or were in the process of partnering. Can you announce who you're partnering with, or not yet?
I can, but not now. Give us a few months. But yeah, we are working with a partner in that space. And we think that they are on the very forefront of protein folding, and we look forward to sharing more in the coming months. Sorry.
Hold on.
All right. May I know the difficulty in building virtual cells?
Is it hard?
The most difficult things in building such a model?
I don't know yet because we haven't finished. What I would say is that one of the challenges today is that I think we and a small number of other companies have gotten really good at building foundation models for specific areas, so like a predictive admin foundation model or a protein-ligand interaction foundation model. That's become pretty straightforward for us with the right data and the right compute. Building multimodal models that integrate many different kinds of data and many different kinds of questions that are more generalizable, I think there's a lot of precedence for how to do that in the text space. There's less precedence for how to do that in kind of a general science space, but that's something that we're working on. I think that's likely to be the biggest challenge that we face is sort of building a single model.
Maybe just picking up on the virtual cell model. I mean, there are different cell types, right?
Yeah.
So maybe just talk a little bit about sort of the dynamism that might maybe going into the model. Can you shift it to kind of reflect different cell phenotypes, cell types?
Yeah, fantastic. So at the atomistic layer, I think that's going to be sort of a pan model. Probably the protein layer will be pan as well, unless there's kind of different proteins being expressed, but we should know that from the pathway level. And I think Recursion today has built whole genome array CRISPR knockouts in more than a dozen completely unique cell types. And I'm unaware of anyone else who's done that at the omics level. Now, that doesn't mean we can predict what all cell types will be, but we're not done building whole genome array CRISPR knockouts. We're going to build dozens more.
And so if you imagine at the pathway level, sort of where cell context is most unique, if we can be the leaders there with data that underlies those predictions, and then lever into that a protein and atomistic model that are more generalizable, I think we'll be in a very, very good place. And we have not only gene knockouts in these contexts, but for example, in some of our inflammatory disease, we've added thousands of cytokines and secreted factors and proteins. And so I think we're getting a pretty, I mean, biology is so diverse. I can't say that it's going to be completely generalizable, but we are getting a very diverse set of input data. And these AI models are tending towards being pretty good at, if you have a sparsely filled matrix of perturbations, at predicting what's missing.
So we should be in a good position there.
Over here again.
Sorry. Follow-up question to what you just asked.
I think it's on.
Follow-up question to what you just asked. Is there a specific cell type that you're starting with for your virtual cell model?
We have the most data in human umbilical vein endothelial cells. I came out of a vascular biology lab with Dean Li, and so I was the first robot and used my favorite cell type. And that's continued to be a workhorse for us. So I think we have the most data there. That said, we now have some really interesting data sets across, for example, with our colleagues at Genentech for different contexts in GI oncology, or perhaps in the near future, multiple different neural contexts in our neuroscience work. And so you'll start to see us build maps that are not only across different kinds of cells from a tissue perspective, endoderm, mesoderm, ectoderm, etc., but even within those areas, starting to vary, for example, mutant cell lines within a specific context.
Again, we won't expect to have to build every possible cellular context in order to get to a place where we could still predict how changing or mutating one specific gene could update a context.
Thank you.
Yeah.
Just as it relates to the partnering initiatives on this particular front, are there certain, I guess, to what extent are just how to think about sort of the scope of treatment areas that are in play? Are there any areas that we'd want to retain or kind of off-limits for the purposes of building out your own pipeline, or are you kind of agnostic?
No, it's a great question, so today, we focus our internal pipeline primarily in rare disease and precision oncology. I think those will continue to be areas where we build our own pipeline. That said, we will very frequently out-license or co-dev programs that are moving forward because we have to find a way to advance a pretty large pipeline. Where we're partnering today: neuroscience, inflammation immunology, large intractable areas of biology, so for example, cardiovascular metabolism, we don't have a partnership there, but you could imagine that's a place we would add one if we found the right partner, a visionary partner, and I think if we are able to go after those large intractable areas of biology with large pharma or large biotech partners while building our own pipeline, it's not just about the medicines that are advancing. It's about all the learnings that are happening.
Those learnings go both ways. A great example of this is we announced at JPM last year this thing called LOWE. It's a large language model orchestrated workflow engine, so it's basically like a ChatGPT for scientists, but it's actually working with our lab and working with our data and allowing you to use AI tools for generative chemistry without being a coder. A couple of months later, we announced that that was being rolled into our collaboration with Bayer. The team at Bayer is now using LOWE with our scientists to collaborate on all of the programs that are advancing through that pipeline together, so I think we'll find ways to learn and to teach with our large pharma partners. We won't do a partnership if it's not like that. We want learning and teaching going both ways.
Okay, great. We'll have to leave it there for time. So thanks very much, Chris. Thanks to everybody for your questions also.