Good morning, everyone. I'm Sean Laaman, Head of U.S. MidCap Biotech Equity Research at Morgan Stanley, and welcome to our Global Healthcare Conference. Before we commence, I'll make you aware of some important disclosures. For those disclosures, please go to the morganstanley.com/researchdisclosures website. If you have any questions, please reach out to your Morgan Stanley sales representative. Post that, we now welcome Recursion Pharmaceuticals with CEO and President, Najat Khan, and CFO, Ben Taylor. Welcome to the both of you.
Thank you.
Maybe do me a favor here and answer sort of three macro-type questions. Just thinking about the rise of China-originated innovation, does it change your competitive positioning and your R&D and BD playbook?
Yeah. Thank you so much, Sean, for the question, and it's great to be here. Look, I would say it, in fact, reinforces what we do. Recursion started with the whole concept of better understanding and decoding biology, so a huge amount of focus for what we do is around novel biology. I was, in fact, just recently in Shanghai two weeks ago, and just the speed and the execution and the prowess in that in optimizing known biology was really impressive to see.
It's also reflected by all of the deals that have been done to date. But that next frontier has to be new and novel biology, really first-in-class targets. That's evidenced by what we have in our clinical program, REC-4881, REC-1245.
We'll talk a little bit more about that, I'm sure, which are both first-in-class targets that came from our engine, novel biology, or novel biology linked to the right indication, which is the case for REC-4881. The other thing I'll also say, we recently announced an unexplored new target that was optioned in neuroscience by Genentech.
That comes from our proprietary data mass that we've generated. Taken together, that's the premise of what Recursion was really founded on around novel biology, and I think it becomes even more important, and some could say prescient, in terms of what's happening with the optimization and the speed in China as well.
Sure. Thank you. I'll skip the next question because it's AI related, but you are an AI-derived company. Which policy variable, is it FDA, Medicare negotiation, MFN, tariffs, or global pricing, do you think about the most, and what matters most to your business on the regulatory front?
Yeah. It's a great question, Sean. I think all of the above are, of course, important to take a drug and actually commercialize it ensures there's access to patients. For the stage of the company that Recursion is, FDA regulatory guidance, et cetera, is critically important, and we have a very productive relationship with regulators. I would say in addition to that, as we think about the commercial potential of our assets, of course, the other policies that you mentioned, MFN and others, are also critical.
Sure. Thank you. Some questions here on strategy. You've been elevated to CEO over the last 12 months. You've cut operating expenses materially. A bit of level setting, how should investors view Recursion now?
Yeah, I know. Nine months, Sean. Nine months.
Yeah.
Counting. I'm just kidding.
Yeah.
Here's what I would say. Recursion is a fundamentally different company than 12 - 18 months ago. Recursion has transitioned from a company with a promise in the novel biology platform to a company where the engine, I like to call it, has been expanded to all parts that you need to make a drug, biology, chemistry, clinical.
To do that, what we have done is we are building the body of evidence in the clinic and beyond in terms of differentiated outputs that are being generated by that engine. I think that is critical. You're starting to see repeatability across the progress we're making, both in our internal pipeline and also in our partnership, and I'll go through that just in a second. The third piece that's materially changed, Sean, as you mentioned, is just our level of productivity.
We have really leaned into AI agents, automation, as well as concentrating capital where we have the highest conviction. We do that every single day. Therefore, we've reduced our cost base about 40% pro forma post the integration with Exscientia. Just to give an example, it's good to say those three pillars, but just to give an example in terms of where we're seeing that body of evidence translating from the engine.
Number one is the first-in-class program in our clinic. REC-4881, first in class for FAP. Total addressable market, just to remind everyone, nothing approved, this is an oral drug, is over $10 billion. We also have multiple programs with one, for instance, REC-1245, RBM39, first-in-class target identified in our platform, the degrader made on our platform, and should have more data later this year. That's just one.
Then we've also made a lot of progress with our partners. We've just crossed over $525 million in upfront and milestones where we're actually advancing first-in-class or intractable biology programs with our partners. These are not isolated events.
I think that's really important to understand. You're seeing this repeatability across from novel biology, novel programs, to the clinic, and that's all produced by one recursive learning platform that we've been building over a decade. The fact that we're doing that with better productivity, as I mentioned, 40% less of our cost base, I think it's a testament to if you get better at learning, you should be able to predict more going forward and be able to validate that in the clinic.
In a nutshell, I will say the company has really transformed to a company with differentiated assets powered by an end-to-end AI engine with repeatability that's coming through both our internal pipeline and our partnerships.
Thank you. How should investors think about the key points of differentiation of your platform, and what do you think investors misunderstand or miss the most?
I think what investors haven't fully captured, I would say, is two things, and it's because of this transformation that's happened in the last 12 - 18 months, Sean, as I mentioned. First is actually the inherent value of the assets itself. I mentioned we have multiple programs, but two that are first-in-class programs. The first one with really high-quality proof-of-concept data already delivered. I think that is underappreciated today because often people are associating Recursion with the pipeline from a couple of years ago. I think that's one.
The second, when you look at the progress across the pipeline and the partnerships, like I said, these are not disconnected. There's a body of evidence collectively that's being generated, and I think people are underestimating how much the platform has changed. It's not a phenomics platform. When I first joined Recursion, people would say, "Oh, yeah, you do some phenomics." Absolutely not. It has expanded from that to multimodal data, where we can find novel targets validated by what we just did with Genentech and what we have in the clinic.
We can make novel molecules, and we have evidence in the clinic, and I think that's a very different engine than what was around 12 - 18 months ago. I would always look for that repeatability, because if you have an engine, I want to know that you can do that over and over again. I think those two pieces are underappreciated, and we have a lot of exciting catalysts coming up to dig into that more.
Awesome. I'll move on to REC-4881 and FAP, but-
Yeah
maybe to give some investors tying the platform and the discovery piece together. How was the platform used to discover REC-4881, and what are some of the properties of the pipeline candidate that make you most excited?
Yeah. Thank you, Sean. This is REC-4881. This is our first-in-class oral asset in FAP. Just to answer the first question, this came, as I like to call our platform, there's three components, the biology, design, and clinical AI.
This came from the biology part of the platform. It has been known for 50 years that FAP is driven by a loss of function in this gene, APC. Nobody had ever studied, and nobody had ever made the unobvious insight that a MEK1/2 inhibitor could actually reverse the challenges caused by APC loss of function. That arc was not known. Nobody had tested it clinically.
What we did is we looked at cells, and we knocked out APC. That's the root cause of the disease. Then using our platform, we screened thousands of compounds to say which compound can actually phenotypically reverse, this is the cell morphology, reverse the cells from disease state to healthy state. This is where we use ML models. The top of the list of predictions from our foundation models was this asset, which happened to be a MEK1/2 inhibitor.
We did not go in with the hypothesis. Think about how you do discovery today. You have a hypothesis and a mechanism, and you go test it out. Here's an unbiased going from disease to healthy, and that's the mechanism that came up at the top of the list. This is where it's really important, the last question you asked me. We generated our own data.
We have our own data factory in Salt Lake City. Super important as you think about LLMs, frontier labs. Everyone's talking about models being commoditized. What's the next frontier? Having your own high-quality data. The next step that's really important is actually testing that in the labs. We did our preclinical model. It looked good with that asset. Then we did our healthy volunteer. It looked good. As Sean just mentioned, we did our proof of concept study.
We saw rapid and durable polyp burden reduction. Rapid in three months, almost half the polyps are gone, and what was even more remarkable, especially for KOLs, is when they are off drug for three months, the effect not only persists but deepens in some patients. Remember, FAP is a chronic disease. It starts when you are 10, 11, even younger, and you are living with this disease for the rest of your life.
It is the most penetrant GI cancer compared to, you can compare to BRCA1 and BRCA2 in breast cancer, 50,000 patients, U.S. and EU5. Standard of care today is surgery, surveillance and surgery. Patients have more than 50%-60% of their GI anatomy removed. Imagine losing that. Your colon is removed, parts of your upper GI is removed, your pancreas, your gallbladder. It is incredibly invasive.
If you can get a disease-modifying drug that can slow that progression of the disease so that you can actually reduce the surgeries, reduce the polypectomies, all of the invasive procedures, that would be incredibly meaningful for patients. Last thing is there is zero approved drugs. Ours, just to summarize, rapid reduction in polyps, which is, and by the way, every polyp is precancerous.
That is the root cause of the disease. Durable, even when the patients are off drug for three months. Oral, which we all know when you get a drug that is oral for patients, it is chronic. The last thing I did not mention, which has been a really exciting piece that we are learning more about, is we see polyp burden reduction in the upper GI and in the lower GI. Let me just take a second to answer why that is important.
Even after your colon is removed, the polyps keep growing, upper GI and lower GI. Upper GI is where you have your other organs, pancreas, gallbladder, and so forth. It is really challenging because that is where you end up getting the invasive surgeries. Another thing about the anatomy of the upper GI, is thin. Which means when you remove polyps, the risk of perforation and bleeding is higher. We are the only investigational asset to date that has shown such rapid, durable polyp reduction and also in the upper GI. So that is really meaningful from an unmet need perspective.
Thank you. You are in active FDA engagement to define a registrational pathway with updated data at CGA-IGC , and regulatory clarity expected this half. Can you provide more color on what we should look for in the second half for those updates?
Yes, great question. Sean mentioned there are two important aspects. One is we are sharing additional information at the CGA-IGC meeting in early November. This is the main congress for inherited GI diseases. This is essentially our target audience for FAP. We are doing a presidential plenary to share additional information on our phase II data, just some of the data I mentioned, but more information. That is number one.
Number two, we are also in active conversations with the FDA. Just to remind everyone, we have Orphan Drug designation, and we have Fast Track designation as well. It has been a very productive set of conversations with regulators. Our base case is essentially what you see [inaudible]-like progression-free survival. What does that entail? These are the meaningful clinical events that you would want to slow down. Surgeries, dysplasia, and so forth. That is what we are focused in our conversations with the FDA, and more to come.
Sure. Thank you. There has been a history of toxicity with MEK1/2 inhibitors. Maybe talk us through what you are seeing on the safety profile. Also, can we touch on the durability off treatment raises the possibility of intermittent dosing. Can you talk us through how you are thinking about dosing? Is that potentially part of the registrational package?
Two questions. Let's talk about the safety profile. It's a MEK1/2 inhibitor. Over 85% of the treatment-related AEs we've seen so far are Grade 1/2. The main ones that we see is dermatitis, which is very much in line with what you see with MEK1/2 inhibitors. We've seen some transient CPK changes and so forth. Just to the question Sean asked, there are actually MEK1/2 inhibitors already approved in oncology, but there's also precedence in rare diseases.
It's approved in NF1, which is a rare disease for pediatrics and for adults. From that perspective, I would say what we have seen is this is very much in line. The dermatitis is something that we are also managing proactively with prophylactic solutions, which has really helped. I would say even in the real world, most of the CPK has been quite transient or even LVEF.
It's very well known in terms of the profile given both in onc and also in rare diseases. That's something we take very seriously. The other thing I just forgot to note is ours is a QD, and we will also be looking at intermittent dosing, just given what we have seen with the durability of this drug.
Some of the other MEK1/2 inhibitors that exist are primarily BID. Again, if you think about it from a patient perspective, a QD, especially for a chronic dosing, I do take something on a QD basis, and I can tell you even that my adherence isn't always the best. That's a little bit around the safety profile. Your next question was focused on-
On the durability of response and-
Great. Yes.
Yeah.
On the durability, we do not know exactly why, but we have some hypotheses. In terms of we are doing some preclinical work. When you look at polyps, is there potential that a MEK1/2 is actually evolving both the composition of the polyp as well as the stroma, so the microenvironment of the polyp as well? There is some translational work we are doing in-house, but I will say what is really interesting from an MOA perspective for this MEK1/2 inhibitor, it acts via a dual mechanism.
One is, of course, the MAP kinase pathway it suppresses. We call this the evolution pathway for polyps, because as polyps mature, you do not just get the APC mutation, you get KRAS, BRAF, and additional ones. The dual part is the second part is it also has crosstalk with beta-catenin. We think that is the other reason that it is also effective.
It has a dual hit in terms of on the polyps, and then some of the work that we are doing in terms of, is there potentially it changing the composition of the cells in the polyp itself and the stromal area? More to come. Look, what we are very excited about and many KOLs, I think this is one of the aspects they found really compelling is that durability when the patients not just durability off drug. Nobody has tested it off drug, and in some cases even deepening.
Sure. Thank you. Moving on to REC-1245 and the broader pipeline. REC-1245 targets RBM39, which came out of your maps. How was the molecule discovered?
Yeah. The discovery of the molecule came from the same maps that I just mentioned for REC-4881. This is what I mean, two non-obvious insights from the same engine. It starts to get interesting. Just recently, we had a new target option by Genentech from the same type of engine. This is what I mean, that you start to see the repeatability. Let me just explain for REC-1245.
At Recursion in Salt Lake City, we have a large data factory where we can manufacture millions to billions to trillions of cells, and we knock out every single gene in those cells, so 20,000. You can measure not just genomics, transcriptomics, and then for validation, we add proteomics and other assays to the list. Very different from the Recursion 12 - 18 months ago. For REC-1245, we were looking at CDK12.
I'm sure many of you know that CDK12 has been a target of interest for some time for DDR modulation and so forth, but it's been hard to drug because of the similarity to CDK13. We looked at what are other potential entry points by looking at all of the other genes that we knocked out that have potential shared biology to CDK12. That's when RBM39 came up.
That was a non-obvious insight. The next thing that we did was we rapidly, within 18 months, designed a novel degrader, and then we went into the clinic. By the way, there's two big pathways. I like to call this asset potentially multiple doors that you can open. One is the genomically unstable tumors. Think about heavily pre-treated patients with solid tumors that are genomically unstable. The other is pediatric transcriptionally driven tumors, like Ewing sarcoma, for instance.
The beauty of it is, I think there's a potential, and we'll let the data drive it for us. Monotherapy, if we see increasingly high dependency on RBM39, you can actually put those cells over the edge monotherapy or potentially combo. Think about the PARP PRMT5. There's multiple options. That's why the addressable patient population, about 100,000 patients, is quite high.
But this is a novel target. Same engine that I talked about REC-4881, same engine that's creating the novel targets for Genentech, and the design platform that made this molecule, same engine as what we're doing with Sanofi, where we've gotten five milestones to date, another important one coming up in I&I. That's how we had discovered the target. At that point, there was not this connection between RBM39 and CDK12 and the pathway and mechanistic understanding.
It's not just a novel target. Where Recursion is going is once you understand the target, you actually get a better understanding of the patient population you'd want to enrich. That's powerful before you dose a single patient. We've been talking about it for the long time, but we're making some good headway there, but we have more data coming on that later this year.
Sure. You've got other programs, multiple programs advancing.
Yeah
How does the data set look or the news flow look over, say, next six, 12, and 24 months?
Yeah. We have multiple other programs. For instance, we just initiated our PI3Kα H1047R program at phase I. We have IND-enabling programs in the NPP1. We have LSD. I think the way, if I step back, Sean, I would think about it as we are constantly looking at our programs. Because we have a repeatable engine, we have multiple shots on goal.
Like we did last year, we continue to look at the programs and say, "Where do we have the highest conviction?" and double down. We already have some great examples to pick from the multiple programs that are coming up. All of these programs have some sort of catalyst coming up in the next year, 18 months.
Sure. Thank you. Moving on to the partnership. Genentech has advanced the first neuroscience target from the collaboration.
Yeah.
Previously unexplored target into joint small molecule discovery. What does an option mean economically, and what triggers the next one?
I will start, and Ben, if you want to add to this as well. I think for us, as Sean mentioned, our partnership with Genentech is focused in neuroscience and in GI oncology, where it starts by generating these novel maps of biology, which Genentech has optioned two already for about $30 million each. Then the next step is how can we take this rich, novel, proprietary data, predict which targets might be causal, and then validate them back in the lab?
The first of those was actually just accepted, as Sean mentioned, by Genentech. One thing to note, for those of you who are watching whether the scaling laws of biology and AI will make a difference, this is one of the first examples of that you can actually find non-obvious biology that is causal from data that nobody else has in the world. But Ben, I do not know if you want to talk about totality of what we have brought in, and also to Sean's point with this milestone.
Absolutely. As you mentioned earlier, we have brought in $525 million in upfront and milestones from our partnerships. We have hit over a dozen milestones. This was another key one because it was also advancing the biology side of the platform.
What we are doing now is moving beyond the maps and actually translating it into a pipeline of neuroscience programs that are all based on completely novel targets. What we have looked at now is the first milestone coming in for that target validation. We will move through small molecule validation. It is important to remember, we run all of our partnership business to be at least break even or profitable from the beginning.
What we are trying to do is really use it not only to build our platform, but also build a lot of NPV value, where we continue to expand the pipeline using the scale and the technologies that we have, then have a really nice profitable back end to all of those programs.
So maybe just two things to note. One is for each program, the milestones are about $300 million, with single-digit royalties for Roche. Not only has the target been optioned, but we are actually working on developing the small molecule for the target. So the journey continues beyond optioning the target. The second thing is we expect potential for more targets from the same data set.
Remember, if you think about the internet corpus of data, that really was available, that is what was used to build a lot of these LLM-based companies. Guess what? That does not exist in biology. That does not exist in science. This is why having a data factory is so differentiated for Recursion.
Not just having a data factory, but having the wet and dry lab connected so you can validate and see if the stuff is causal, then actually design the molecules and take it to the clinic. So the fact that we have connected it end to end, then we have the data factory, means these maps are reusable. That is the point I am trying to make. It is a durable asset because you are mining it over and over again to find novel targets, and we expect to see more of that.
Sure. Maybe just to clarify my mind, so the neuroscience maps are used whole genome CRISPR knockouts in iPSC-derived neurons. Is that data set reusable across other partners, or is it exclusive to Roche?
It's exclusive to Roche. We have a 10-year exclusive neuroscience partnership with Roche. Kudos to Roche, and they're phenomenal partners, just like Sanofi. We don't do volume partnerships. We do very integrated partnerships because, again, we're doing things no one has ever done before. It's not a rinse-and-repeat play. This is a complete frontier play. Yes, it's exclusive with Roche.
I think important note on that. Roche has paid us over $250 million so far for the program.
Yep.
We actually own all of the data. Through that partnership, they only have access to the work product. To Najat's point, we do have another five years on the partnership and hope to continue it. They're a fantastic neuroscience partner. We are actually keeping the in-house capabilities and data that come from it.
Awesome. Thank you, Ben. You notched a fifth Sanofi milestone. How should we think about the Sanofi collaboration going forward? What are you focused on in that collaboration?
Yeah. The Roche/ Genentech partnership, which is focused on novel biology and now moving into programs? For Sanofi, it is focused on developing small molecules for intractable I&I and/or oncology targets. Sean, as you mentioned, the five milestones we have achieved so far, it is for lead series for making those compounds against those intractable and first-in-class I&I and oncology targets.
The next one that we have given guidance, potential one that we have given guidance for, would be a milestone at development candidate for a first-in-class oral small molecule program. That would be the most mature one. I think that as a standalone from a value proposition, it is a highly valuable asset, but there is also a read-through in terms of the design part of our platform. Just to remind everyone, we do not design everything.
Small molecule degraders, that is what we have focused on because we think that the activation energy to figure that out is pretty high, so that is why the competitiveness is high. If that is option a development candidate by Sanofi, which we have given guidance for, that would then enter their pipeline. This is where Ben was saying, then it becomes we do not have operating costs, and we start to recoup the milestones.
Again, just to reiterate, the milestones for Sanofi are over $300 million, and the teen level royalties as well. These are meaningful milestones. I want to underscore, for a company like us, as we are building our engine, I keep talking about everything you have heard about, the Roche targets, the clinical drugs that we have in our pipeline, the Sanofi partnership.
It is all coming from the same engine, and the recursive learning from the engine is incredibly important. These partners have been phenomenally helpful for us to not just have cool data, but to actually learn how to turn the platform into an AI native product or asset engine.
Sure. Thank you. Maybe onto one of my favorite topics, the platform and the AI question.
Yeah.
Yeah. You cite roughly 330 compounds per development candidate against 2,500 industry-wide, and the target to candidate in one and a half years, there seems to be versus forward, seems to be consistency in the industry on that metric. But that's a productivity claim. Can you walk us through the evidence that the molecules are better rather than just cheaper?
Yeah. Look, quality matters the most, and then faster is definitely a nice couple. Here's what I would say, what we just talked about right now. Of the $525 million in upfront and milestones, $125 million of that is in milestones. I think we have one of the highest milestones achieved for any of these AI native drug discovery and development companies.
That matters because upfront is promise, milestones is proof. It is the quality of the molecules, for instance, with Sanofi, for some of the other work that we are doing, that is a signal. The other thing I would also say is the work that we have in our programs that we already talked about. This is what I mean. Look, we are learning as we go, but we are building this body of evidence across biology, chemistry, and the clinic.
Just to reiterate the speed part of it, speed is important because if the thesis is that you can have high-quality data that generates better and more performant predictions, then you have to validate only a few to get to the right answer, right? We believe that is true. You see that in every other industry. Our timelines are about 3x faster for small molecules, which are hard, than industry.
We are physically making about 90% less molecules physically, because the ones that we are physically making are closer to the end point of the desired state. I think that is a velocity in the engine that I would, again, by itself, you do not claim victory. At the end of the day, the drugs have to work in the clinic. But I always say these are green shoots to watch. Anything else you want to add?
Yeah. The only thing I would say is one of the really important elements is even though we use AI, we actually create physical things. You can take all of these chemistries into the lab and test them against other chemistries that have been developed for the same targets.
There is a very quantitative analysis that you can do to say, "Is this exceeding what has traditionally been done or not?" We have had more than a dozen different development candidates, as Najat mentioned. We have had external partners that have been validating it. We are seeing really powerful output from the design platform as well as the biology platform.
I will say, everybody is using some sort of LLM. It gets better the more you use it. That recursive learning is actually really important. Do we get everything right? No. But look at our success rate. In industry, it is 10%. You get to 20%, you double it, right? That recursive learning, it would be hard to do that if you just do novel target biology. Because you have to make the compound, and then you have to go in the clinic, and you have to learn from that really fast.
I think that is where our focus is. Do all of that to unlock a new TAM. That is why the first-in-class programs matter, whether it is our own pipeline or with Sanofi or with Roche, and do it fast so you can learn fast, right? You can start things, double down on things that work, and what does not work, learn from it really fast. That is how we are trying to collapse that engine and the timeline, but the reason is to improve that recursive learning.
Sure. Thank you. [inaudible] just described as the data factory, the lab in the loop.
Yeah
The differentiated pipeline. Frontier models, as you just mentioned, keep improving and available to everyone. Which of those three is the hardest to replicate?
Look, I think the hardest to replicate is actually integrating it together and having an end-to-end engine. There are so many steps that go into biology all the way to the patient and showing data. I think unless you learn fast, you're not going to bend that probability curve that we all want to bend. But in order to do that, Sean, I would say the models are great.
Model architecture is important. But unless you have the right data to teach it, the predictions go back into the physical world to experiment and learn from it, what's working versus not, that's, I call it almost the patient in the loop, because not even just the lab, because we're going all the way to the clinic. But the data factory is critical, especially in this space. Because unlike some of LLM and others, LLMs are not going to be sufficient for science.
You do need other modalities of data. And we have proprietary protocols for the cells that we're making. We make 1 trillion cells, iPSC-derived neuronal cells. I used to be in lab, didn't even want to make 100. It's super hard. I used to just work with HeLa and CHO cells. These were so much easier.
The level of sophistication in our science data factory, the amount of automation that it takes, we've been at it for 10 years. There's something to be said about that, because you really become much more mature and seasoned on what works versus not. And now we have the output to show for it, not just datasets b ut actually novel targets being optioned by really high-quality research organizations.
Well, thank you, Najat. We've just run out of time, so thank you, Najat. Thank you, Ben.
Thank you.
I appreciate you participating.
Thank you, Sean.
Thank you.
Thank you so much.