Recursion Pharmaceuticals, Inc. (RXRX) Earnings Call Transcript & Summary

Good morning, good afternoon, good evening, everyone. Welcome. I'm Najat Khan, Chief R&D Officer and Chief Commercial Officer at Recursion, and I'm joined by Dave Hallett, Chief Scientific Officer at Recursion. Today, we have the opportunity to share with you the Phase I monotherapy data for our CDK7 program. This is an interim update, and I'm excited to share with you some of the progress the team has made. If we can go to the slides, please. Fantastic. First, I'll start as a reminder that we're going to be making some forward-looking statements in the presentation today. I'll start by giving you a summary of the key highlights from our interim update, and then I'll hand it over to Dave to share some of the details of the data and the program and the path forward. So as a quick summary, Recursion-617 program, this is a potent orally active, highly selective reversible CDK inhibitor. Now you might ask, why CDK7? CDK7 plays a dual role in regulating cell cycle progression and transcription, both of which are often dysregulated in cancer. And we've seen the impact that CDK4/6 inhibitors, which target cell cycle progression have had as effective cancer treatments. CDK7 inhibition has the potential to build on that by offering the potential to target both cell cycle progression as well as transcriptional addiction, which is a hallmark of aggressive tumors. The next question is why REC-617. We have developed 617 to fine-tune and optimize the therapeutic index, focused on increasing permeability, ensuring rapid absorption and limiting GI exposure while also ensuring that there's high selectivity. You'll see some of that data from Dave shortly. And how was that done? This is a program that was developed in partnership with GT Apeiron and was recently licensed by Exscientia before the close of our deal. They worked together using Exscientia's physics-based generative AI and active learning approaches to optimize the molecule design. Now one of the things to note is from hit ID to candidate ID, it only took about a year, less than 140 compounds to design. So where are we today? We're in dose escalation, and this is an interim update. Dave will share some of the data around the PK profile, which is promising, [ ESCs ] which are dose linear and exposures well above the IC80 at peak. We're also seeing rapid absorption part of the design elements and criteria that I discussed briefly. We have also measured PD in this case, using POLR2A [ code ] change, where you see that it corresponded really well to the PK and shows strong target modulation about 80% to 90% at 10 to 20 milligram QD. We will also be sharing some of the interim efficacy and safety data. From a safety endpoint -- from a safety standpoint, 617 is well tolerated, mainly Grade 1 to 2 AEs with no discontinuations due to AEs. And one thing of particular to note, and while this is still an interim update that per the design, we're seeing less GI side effects. On the efficacy front, we are pleased to report a confirmed partial response, where the patient had progressed through 4 lines of treatment prior to starting the study, so a heavily pretreated patient. And this patient has been on the treatment for more than 6 months and ongoing. So we're seeing a durable response. And we're seeing this response across 2 lymph nodes in terms of reductions for RECIST 1.1. In addition to that, we're also seeing 4 patients with stable disease, best response to date and up to 6 months of treatment. I'll remind you that the NPD has not been reached. So a couple of the next steps for us is to continue monotherapy dose escalation in both QD and BID as well as we're excited to initiate combination studies in the first half of 2025. With that, I'm going to hand it to Dave to walk us through some of the details and the data of the program. Dave?

Thank you very much, Najat, for introduction, and welcome to everybody online today. Before I get into the main discussion points, I just wanted to remind the audience of Recursion's design platform and design philosophy as it relates to the precision design of the small molecules. At the start of a project, one never knows what kind of data sets will be most useful during the design process and at what point. And so more than a decade ago, we deliberately set out to design a platform that was data source agnostic. It's capable of utilizing crystal structures, for example, where they exist, but it's equally at home in the absence of crystal structures and even in the arena of phenotypic drug discovery, where at the start of a project often we don't know actually what the protein target actually is. We also don't know what kind of statistical analysis method is going to be most pertinent. Sometimes it might be, again, physics-based, so using quantum mechanics or molecular dynamics when one has a crystal structure. But it could equally be a multi-task linear model or even kind of Bayesian. Our design platform then leverages generative design and often on each given design cycle may kind of produce in silico tens of thousands, if not hundreds of thousands of molecules, which is scored against reward function that we call the target candidate profile. These algorithms actually produce molecules that are actually synthesis accessible. So we have a synthesis sympathetic kind of component to this. What drives the efficiency, though, is the use of what we call active learning. This is a modern kind of computational method using information theory that selects those -- the smallest subset of molecules, typically 10 to 15 that if synthesized and tested will provide the most information gain back to the system, so trying to drive the project forward whilst improving the underlying models. We then take those molecules onto our automated synthesis platform that we've made significant investment in over the last few years. And the whole idea and the philosophy is about how to most rapidly drive this design, make, test, learn cycle. And in the case of 617, as I pointed out, is that we're able to design this molecule using kind of modern machine learning techniques in just 10 design cycles over the course of just under a calendar year and with just 136 compounds. So this is a significant improvement on both capital and time efficiency compared to modern drug discovery. Just a little reminder about kind of why CDK7. If you look at the various protein targets that CDK7 phosphorylates in the cell, there's a lot of attributes of this particular protein that are relevant to cancer. So it phosphorylates components of the cell cycle and indeed, inhibition of CDK7 typically causes cell cycle arrest in the G1 phase. It also phosphorylates the C-terminal domain of RNA polymerase 2, and inhibition is known to be beneficial in terms of suppressing transcription and particularly suppression of MYC as well. Moreover, cell lines that are resistant to kind of typical standard of care in breast cancer these days, the 4/6 inhibitors, they still respond to CDK7 inhibition. I think that's because CDK7 in some ways, can be thought of being kind of upstream of CDK4/6. And again, in the context of hormone receptor positive cancers, CDK7 actually phosphorylates the H2-receptor itself and so has relevance in that setting. So the idea at the start of this journey was that if we're able to combine a CDK7 inhibitor with orthogonal agents targeting complementary pathways, we may be able to achieve a more comprehensive antitumor response than a single agent or standard of care today. But here's the problem. CDK7 is indispensable for cell proliferation. So any healthy high turnover cells. So for example, in cells that line the GI tract, the epithelial cells or even certain kind of blood cells, they need to be managed. So how do we go about thinking about this? How do we get a therapeutic index greater than 1? This is a kinase. So we need to be highly selective for CDK7. First-generation molecules were typically nonselective for the wider CDK family and significant systemic toxicity resulted from that. Because of the potential I mentioned about trying to avoid the -- or spare the epithelial cells of the GI tract, we wanted a compound after oral dosing that was rapidly absorbed. So it need to be highly permeable, and we wanted to minimize kind of interactions with transporters that are often found in the GI tract and even in the tumor microenvironment. We wanted a reversible mechanism of action to enable the fine tuning, the control of the target engagement. A number of CDK7 inhibitors are irreversible in nature. And the problem with that is that you have to wait for the protein to be resynthesized before its normal function and returns. We wanted to manage the time on the target while limiting drug holidays. CDK7 is, I think, is particularly relevant for highly aggressive tumors. And I can guarantee you that tumor is not taking a holiday while the patient is off drug. We also wanted a short human half-life to enable the clinical team to optimize dosing regimens. Remember, we're trying to help manage therapeutic index here. Some of our peer group of half-lives in the 2- to 3-day region, which we just felt was too long. Those first 5 bullet points talk about sort of physical properties. Clearly, another way that we can kind of help to manage therapeutic index is to select those tumors most likely to respond. So those tumors have a high dependency on CDK7. So a little bit of preclinical data just to set the scene. So what we're looking at here is a CDX model. So this is an OVCAR3 cell line. It's hormone receptor positive and cyclically amplified. And we're looking at here are 3 oral doses compared to vehicle. Important to note that none of the dosing arms do we see any significant body weight loss. This kind of these doses were well tolerated. And importantly, at 10 milligrams per kilogram once a day orally, we saw complete regression in all the mice at day 27. We've actually published very similar data in a triple-negative breast cancer CDX model 2, where, again, 10 milligrams per kilogram was the optimal dose. So what is the -- at those particular doses, what is the time course of exposure and kind of target coverage look like? And so this is nicely shown here on this graph. So what we're seeing here is a 10-milligram per kilogram oral dose in these mice. What you can see is that probably for around 8 to 10 hours of a 24-hour period, we're at or above the CDK7 IC80. Pleasingly, we have a good separation for CDK2, which is the closest off target that we could find during cancer screening, so well below both the IC80 and the IC50 even at Tmax. And the idea here is like, okay, is this a profile that we could kind of replicate in the clinic? Can we give good target coverage for the first part of the day, but then allow the function of CDK7 to return and therefore, managing the best balance between safety and efficacy. One of the beauties of our platform is that it delivers well-balanced compounds. So this kind of traffic light chart or stop light chart that we use kind of typically in our presentations compares 617 with 2 orally delivered kind of peer compounds. And what you can see here is that those peer groups, they have some kind of positive attributes. So in one case, potent CDK7 and nicely selective. But neither of those compounds is optimal in all respects. In particular, both compounds suffer from very poor passive permeability and a substantial substrate for transporters. And we believe that our compound should deliver high permeability with rapid absorption in the clinic, whereas the 2 peer group compounds think will take a while to be absorbed. You'll see high local levels in the GI tract, and that may contribute to the GI-related anti adverse events. So on to the clinical study. So we're currently in the monotherapy dose escalation phase. So this is a group of patients that have progressed following or are intolerant to available standard of care treatments. We currently have got as far as 20 milligrams once a day in one arm, and we recently started 1 milligram BID in the second arm and are actually expanding in both those. Remember, the outputs of this are looking at pharmacokinetics, pharmacodynamics. It's important to remember that we haven't reached MTD yet. It's also important to remember that in this particular protocol, this is the first time that we were evaluating this compound in humans, we chose not to offer prophylaxis at the start of treatment for nausea, vomiting or diarrhea. And that's because we want to see kind of directly how well this compound was tolerated. Clearly, in subsequent clinical trials, we can revisit this in terms of how to better manage potential AEs. Another -- here's a patient kind of population so far. So 19 patients in total, 8 of which are response evaluable. Median age of about 60 years, roughly 50-50 male, female. And the 5 tumor types are kind of shown in this kind of table. The important takeaway from this slide is just the number of prior treatments that these people have received. In most cases, the tumors were actually become metastatic and they spread to different parts of the body. So a really kind of difficult kind of patient cohort to evaluate in this Phase I setting. So I'll take you through the pharmacokinetics, pharmacodynamics first. Pleasingly, and as designed, we see very dose linear pharmacokinetics, a very rapid absorption. So Tmax typically around about an hour and as designed, a short half-life of about 5 to 6 hours. And that gives you the exposure profile shown on the left. So again, at both 10 and 20 milligrams once a day, we see good coverage certainly of IC50 or certainly IC84 a portion of the day. But again, decreasing kind of over the course of 24 hours to essentially kind of 0 levels. Compare that on the right with -- this is looking at message for POL2RA, a really robust target engagement, which is dose dependent at 20 milligrams, we see typically 3x to 4x the modulation compared to baseline. And that roughly corresponds from our analysis to about 80% to 90% target engagement. So we're hitting this target very hard for part of the day, allowing the kind of the target to recover during the remainder of the day as we designed. Just to point out, not shown here, we continue to evaluate twice-a-day dosing, and we'll update during a medical conference in due course. How does our molecule compare to the peer group that are also being dosed orally. So what we're showing on this particular plot is the [indiscernible] molecule at its MTD, that's 3 milligrams kind of once a day. And we're showing the steady state levels of samuraciclib at 360 milligrams once a day. And samu has a half-life in humans of about 75 hours and so wanted to show steady state concentrations. And again, we showed 20 milligrams of 617. And what you can see is that 617 as well as being much more rapidly absorbed and then declining over that 24-hour period has a significantly higher coverage of the CDK7 target. And we're optimistic this kind of short -- these design parameters, including a shorter half-life give us the right balance of superior target coverage, but also sparing of the CDK7 mechanism too. So what about the preliminary safety data? And as Najat pointed out at the start is that the adverse events that we've seen so far were predominantly low grade and importantly, they were reversible upon treatment cessation. It's important to note that so far, we've had no treatment discontinuations due to AEs. We've had a very low drug-related rate of diarrhea compared to competition. Again, remembering that this is largely in the absence of prophylactic use of antiemetics or antidiarrheals. We've seen 3 treatment-related adverse events so far in 2 of the patients, and they're highlighted on this particular slide. Again, important to note that nobody came off to the drug is that the event resolved and treatment continued after a dose reduction. Now we weren't expecting to see any efficacy in this early study because of the challenges of the kind of patient cohort we're looking at in terms of their prior treatment lines. But we're delighted to kind of report some initial responses. So we saw a confirmed durable partial response in an ovarian cancer patient. So the 69-year-old female that have been diagnosed with Stage III ovarian cancer back in 2019, she had platinum-resistant ovarian cancer and have progressed following 4 lines of prior therapy and had metastatic disease to both the lungs and the lymph nodes. A particularly kind of challenging patient in this regard because there are no BRCA mutations, very low tumor mutational burden, small TP53 variant allele frequency as well. We initiated therapy in this subject at 20 milligrams once a day. We dose reduced at week 4 due to transient grade 3 nausea and that patient has stayed on and is still on 10 milligrams once a day. In addition to this 1 subject, we had 4 additional patients in different tumor types that also achieved durable, so up to 6 months of treatment responses of stable disease as a best response across multiple dose levels. Important to note that all 4 of those patients have progressed prior to enter the study, 3 of them are in very late stage kind of sixth, seventh line colorectal cancer and one was a fourth-line non-small cell lung cancer patient. One of those patients was on 2 milligrams once a day and the majority were on 10 milligrams once a day. So a patient little vignette here from that ovarian cancer patient. So what we're looking at here is a CT scan at baseline at week 16. So we're delighted to see a partial response here across 2 lymph nodes and at week 16 with a normalization of an ovarian cancer marker lactate dehydrogenase. Important to note that 2 additional biomarkers are also reduced. So cancer antigen 125 is reduced and is kept down by more than 44% over the course of the study so far. And the thymidine kinase 1 is also significantly reduced by 70% -- almost 70% and maintained in the study. Really important to note here that response is ongoing. This patient is still on drug, still on 10 milligrams once a day and continues without the need for any antiemetic intervention. So where are we going to go next? We're going to continue to finish off the monotherapy dose escalation in both QD and BID settings. We're interested to see if the BID dosing schedule actually might provide a more optimal kind of coverage, we will see. Combination studies are expected to initiate in the second half of next year. And myself or colleagues will give further updates on both elucidate and initial preclinical work we're doing with this exciting molecule. In addition, now that we're part of the Recursion family, we can now leverage Recursion's multimodal real-world data expertise, causal AI models and access to the Tempus data, and I'm looking forward to kind of diving into that. Finally, I got to acknowledge the people that did the work. So the investigators are helping to support the ongoing clinical study. A fantastic group of individuals of Recursion, so the design team. So particularly Olivier, Jeremy and Ross, who formed the core team of design and biology at Recursion. And I'd also like to call out particularly kind of Fred and James and Mingxi at GT Apeiron, who are our early partners with this particular program. And finally, because it's why I do drug discovery, it's why everybody works with Recursion, very much like to thank the patients and their families that are involved in this elucidate study. And with that, I will pause for questions and hand over to Najat.

Thank you, Dave. Thank you so much for sharing the data. And we wanted to also share that there is another update that we have upcoming. In the fall, we had shared data -- initial data from our SYCAMORE study. This is our Phase II randomized placebo-controlled trial in CCM. As a follow-up and as a late breaker, it has been accepted for an oral abstract session at the International Stroke Conference. So we're very much looking forward to having Dr. Jan Burkhardt, who is the Division Head of Cerebrovascular Surgery at UPenn, sharing some of the data from that program, both overall, but then also some really important patient segments where the analysis was further done. This will be shared on February 5, 2025. So stay tuned. And with that, we'll open it up to Q&A.

We're just getting some questions in here. Let me see. All right. We'll get started. We have quite a few questions around, can you tell us more about the PR patient and the patients that had stable disease. Can you add some additional details on the patient profile as well as their characteristics? Dave?

Yes. Maybe I'll first touch upon the ovarian cancer patient. We've given quite a lot of information during this presentation. Something I'll remind the audience of though, remember, back in 2022, we presented a poster at ENA, where part of that work is that we took malignant ascites from ovarian cancer patients. So probably a very relevant kind of sample that probably represents both the primary and the potential metastatic phase of ovarian cancer. What we also kind of show in that study, again, using primary tissue is that the samples kind of broke into 2 response cohorts. Around about 60% of the samples responded really, really well to 617 and approximately 30% to 40% required higher doses to see effect. When we chase down the molecular signature and asked the question, why do we see that differential response, again, thinking about precision medicine and targeting patients. It turns out that the high responders correspond to high-grade serious ovarian cancer. So while I don't want to speculate today about the kind of particular demographic of that ovarian cancer patient, I find it personally intriguing that, that data kind of that was generated almost 2 years ago, I think highlighted the potential use of this agent in ovarian cancer. As we kind of pointed out, I think, at the start of this presentation, I think 617 and CDK7 itself have potential in a lot of different avenues, particularly very aggressive cancers.

Thank you, Dave. And just maybe to build on a point that you made earlier in the presentation that's really important. These patients were heavily pretreated. So when we look at the stable disease patients, I mean, sixth, seventh line in CRC, the median PFS for those patients are quite low. And these patients, stable diseases respond up to 6 months. And usually, the median PFS is around 2 to 3 months. So I think that's encouraging for us. Again, we did not expect this given this is a monotherapy. There's a lot to be done. We're going to do a lot of work in terms of really fine-tuning our patient stratification strategy, leveraging our causal AI work that we do at Recursion and clinical genomic data, including the data that we have from Tempus. So a lot to be done and really doing high unmet need for patients. All right. I'm going to go to the next question. This is from Vikram at Morgan Stanley. What level of read-through do you think these initial data provide to help validate the Exscientia-related drug design portion of the current Recursion platform? I'll start and then Dave, I'd love for you to share a bit more. I mean I -- when we look at the Exscientia, legacy Exscientia platform, really focusing on precision design, starting from using physics-based approaches like QM/MD, using generative AI to really create novelty in the molecules that are being developed and then very importantly, active learning in order to ensure that there is optimal multiparameter optimization to get to the right candidate. And you see that happening. You heard today, 136 compounds from hit to candidate ID. That's not a one-off. And you'll hear this time and time again in terms of some of the other programs, whether it's MAD1 LSD as well as you've heard for RBM39 from Recursion -- legacy Recursion program that was also done in under 200 compounds synthesized. That's a significant shift from what we see in the industry, both from a timing perspective and the number of compounds synthesized. So 200 versus 2,500 or more is usually the benchmark and the timing being under a year. So I think that speaks to the ability to get to the right answer better and faster using these approaches. And we're very excited to combine this with the approaches of Recursion, focusing on the leverage -- leveraging AI and multimodal data to really understand not just a biological signature, but increasingly also the right patient for the right treatment at the right time. So harnessing all of that is how we will drive precision medicine even more effectively, and you're starting to see some of the green shoots of this in the programs that we have shared, but much more to come. Dave?

Yes. I just -- I think for those people that follow Recursion and followed Exscientia, I think a couple of times, Ben, our CFO and myself were asked kind of within the year, what would good look like out of the study? And Ben and I reminded that audience of, look, CDK7 is a therapeutic index story. It's about threading the needle. We believe it's a fantastic target, but it has not really been tested properly in the clinic yet because people didn't put enough thought into the target candidate profile. I think what we've delivered today is kind of proof of that design principle. So we delivered a compound that was designed to be well absorbed, and it is. It was designed to have a modest half-life, and it does -- it was designed to give very robust target coverage, and it does. I'll be honest, given the mutational burden that's probably start in those kind of patients, I don't think any of us were really expecting to see any efficacy, I'll be honest with that. So that is the icing on the cake. But yes, I think it's a testament to the design teams and the platform that we're able to sit here today and share this information.

Yes, translating design to outcomes. I mean it's hard work, but that's how we will change how drug discovery is done. I have another question from Dennis at Jefferies and Brendan of Cowen. Can you talk about how much further you can dose escalate and whether you feel confident in the safety and tolerability that it will still be manageable as you dose higher? Does the data suggest less frequent dosing would provide adequate coverage here? Dave?

I guess time will tell, but it's a great question. We've not reached MTD yet. I think the -- clearly, we are starting to see some adverse events as we get kind of higher in the -- as we get to 10 and 20. But a reminder that the ovarian cancer patient has been taking this molecule -- this drug at 10 milligrams for more than 6 months now. So it's well tolerated at that dose for sure. Whether we can go beyond 20, time will tell. Remember, the hypothesis I laid out at the start is that the kind of the data we have in kind of mice would suggest that kind of covering the target effectively for about 8 to 10 hours a day and then releasing it is certainly one way to kind of achieve maximal efficacy. But the BID dosing regimen from simulations looked really, really interesting. I honestly don't know. I'm just really excited to see how the clinical kind of team progresses through that BID dosing regimen. It's a testament to the design team that we even do this. Remember, if we come away, say, a 75-hour half-life or a 24-hour half-life, we could not have done twice a day dosing. So again, I think that the design team did a wonderful job in delivering an agent that allows the clinical team some flexibility as they work through different dosing regimens.

Thank you, Dave. Yes, more to come about QD, BID dosing, and we are excited to share more of the information with you as it evolves and also excited to advance the program into combination studies in the first half of 2025. So with that, I'm going to thank everybody for taking the time to listen in on some of the latest and greatest out of Recursion today and looking forward to a follow-up conversation soon. Thank you, everyone. Thank you, Dave.

Thank you, everybody. Thank you, Najat.

Thank you.