[First of two parts]
For years, drug developers that apply artificial intelligence (AI) have touted the benefits of the technology—the prospect of speedier development of new treatments at lower cost, thus reducing the sky-high price of medicine for patients—while often bristling when questioned by analysts, reporters, and others whether their clinical activity is more hype than hope.
However, Recursion, a leader in AI-based drug development, has met the hype-or-hope challenge head-on, by going public since June with plans to announce clinical data from at least seven pipeline candidates over the next 18 months—or as many as 10, if the company can complete its deal to combine with AI drug pioneer Exscientia.
Recursion began delivering on its promise Tuesday, when it announced positive topline data form the Phase II SYCAMORE trial (NCT05085561) showing that its symptomatic cerebral cavernous malformation (CCM) candidate REC-994 met the study’s primary endpoint of safety and tolerability. Recursion also said that REC-994 demonstrated “encouraging” trends in objective MRI-based exploratory efficacy measures at the highest dose of 400 mg, specifically reduced lesion volume and hemosiderin ring size in patients vs. placebo.
“These results provide critical insights that will inform our next study design, including exploring study duration, higher doses, and a larger cohort of patients,” Najat Khan, PhD, Recursion’s chief R&D officer and chief commercial officer, said in a statement.
But investors were less impressed since the announced data did not include details on improvements in patient or physician reported outcomes at 12 months (Data is expected to be presented at a conference next year). Recursion shares fell nearly 17% in Tuesday trading, to $6.07 from $7.28 on Friday. “We think there’s limited visibility on the path forward for REC-994 in CCM,” Jefferies equity analyst Dennis Ding wrote Tuesday in a research note.
During the fourth quarter, Recursion plans to announce preliminary Phase II data in neurofibromatosis type 2 (NF2) from a Phase II/III trial (NCT05130866) assessing REC-2282, an oral, central nervous system (CNS) penetrant small molecule histone deacetylase (HDAC) inhibitor designed to treat progressive NF2-mutated meningiomas.
Next year, Recursion plans at least five additional clinical data readouts:
- REC-4881—Announcements of preliminary data from the Phase II TUPELO trial (NCT05552755) in familial adenomatous polyposis (FAP), and from a Phase II trial (NCT06005974) in advanced AXIN1/APC-mutant cancers, both planned for the first half of 2025.
- REC-3964—Preliminary data readout in Clostridioides difficile (difficile) infection by the end of 2025, following the launch of a Phase II study (NCT06536465) in the fourth quarter of this year.
- RNA binding motif protein 39 (RBM39) advanced HR-proficient cancers—Dose-escalation data readout from a Phase I trial by the end of 2025, following an IND submission in the third quarter of this year, followed by Phase I/II initiation in the fourth quarter.
- Target Epsilon (Fibrotic Diseases)—Readout of data from a planned Phase I trial in healthy volunteers planned by the end of 2025, following IND submission in early 2025.
Exscientia announced positive Phase I results in May for EXS4318, a selective protein kinase C (PKC) theta inhibitor designed by Exscientia and in-licensed to BMS under an up to $1.3 billion collaboration. Should Recursion combine with Exscientia, the resulting company plans additional data readouts that would include:
- EXS617, a cyclin dependent kinase 7 (CDK7) inhibitor indicated for transcriptionally addicted cancers. A readout of topline pharmacokinetic, pharmacodynamic and safety monotherapy dose escalation data from the Phase I/II ELUCIDATE trial is planned for the second half of this year. In July, China-based GT Apeiron agreed to sell its partial ownership of EXS617 to Exscientia, giving it full control of the drug.
- EXS74539, a brain penetrant LSD1 inhibitor, for which Exscientia has planned to submit an IND later this year, and initiate a Phase I/II trial in early 2025.
- EXS73565, a MALT1 inhibitor for which Exscientia said last month it remained on track to submit a clinical trial application in the second half of 2024. Exscientia has planned to launch a Phase I/II trial of ‘565 in B-cell malignancies, including chronic lymphocytic leukaemia (CLL), in early 2025.
In an exclusive interview with GEN Edge, Chris Gibson, PhD, Recursion’s co-founder and CEO, recently discussed his company’s growing pipeline, its busy next several months, and its agreement to combine its operations with those of Exscientia. (This interview has been lightly edited for length and clarity; Part II appears separately and can be read here.)
GEN Edge: How and why did Recursion come to focus on cerebral cavernous malformation (CCM) as its first candidate?
Chris Gibson, PhD: I have to take you all the way back to the start of Recursion. ‘994 was a product of my graduate schoolwork, my dissertation. And the discovery of the potential for that molecule in the context of CCM—which is what I was studying in graduate school—gave us some sense that maybe we could expand that early work into a company.
I actually finished my PhD on REC-994 in CCM, where I used open-source software written by Anne Carpenter [PhD, Senior Director of the Imaging Platform, Institute Scientist] at the Broad Institute to do an image-based screen to identify that molecule in Dean Li [MD, PhD]’s lab [Li is now Executive Vice President and President, Merck Research Labs.]. After that work, I said, ‘Well, let’s go start a company and try and do this for many other diseases.’ And after a couple of years of building the platform, we went back to the university and licensed the know-how around REC-994 as one of our first candidates.
The company did the preclinical work, the IND enabling studies, the Phase I. We’re excited now to be about to read out the topline data on the on the Phase II.
GEN Edge: What does Recursion consider the significance of REC-994 in CCM, and of the Phase II trial?
Gibson: We’re the first-in-disease. Nobody had previously ever gone to the FDA in this disease. The [trial’s] primary endpoints are safety and tolerability. The secondary endpoints are exploratory in nature. We were looking for where in this disease we might find a measure that we and the agency could get comfortable with taking forward into a subsequent confirmatory study.
GEN Edge: Why does the Phase II study have primary endpoints of safety and tolerability as opposed to the efficacy endpoint more common in mid-phase studies? Was the FDA looking for that?
Gibson: They leave it up to you. But I think we and they both believed that because nobody had done a study in this disease, [and] because this would be a treatment that people would probably take for a very long time, that the best thing to do was to really focus on safety and tolerability because that’s essential. But too, we wanted to measure lots of different potential efficacy endpoints to try and identify something that we could take into a larger study. So, this is the design that we agreed to and took forward.
GEN Edge: How does REC-994 stand out among other CCM candidates, such as Neurelis’ NRL-1049, Ovid and Graviton’s OV-888 and efforts to add CCM indications for established drugs like atorvastatin and rapamycin?
Gibson: We are the first company to do an institutionally backed clinical trial in this space. I think the biggest differentiator is that we’re several years ahead of any other kind of commercial opportunity. Ovid and those others are maybe beginning Phase I. A number of those other companies also are working on a mechanism related to rho kinase (ROCK), which is the mechanism which failed in my graduate schoolwork, which is what led us to go look for something else like REC-994. Now, that doesn’t mean it would fail in their hands. And there’s differences in the details of the molecules they’re pursuing. But certainly from our perspective, if we’re several years ahead with a mechanism that’s different and highly, highly safe, then that could be very beneficial for patients.
The challenge of being first in disease is, it’s an exciting opportunity to make an impact for patients. It also means that there’s an increased burden to not only prove out the molecule, but the path itself towards the treatment. So, lots more work for us to do, regardless of the outcome. But we’re keeping our fingers crossed for sure.
GEN Edge: Among other candidates in Recursion’s pipeline is REC-3964, the C. difficile candidate.
Gibson: Yeah, that’s our first new chemical entity that takes into account some of our chemistry tools as well. With the older molecules, we used our early ML and AI tools to identify the opportunity of a known small molecule in a new disease area. We licensed the molecule or took an old molecule off the shelves and took it forward. Again, I believe ML and AI tools can be used in lots of ways to improve the way we discover and develop medicines from target discovery to hit discovery, to chemistry, to clinical trial design.
It seems that many believe that the only way to call a medicine an AI medicine is if the chemistry was done with AI. I think that’s just a fallacy. I believe if you get a medicine advanced towards patients that would not have otherwise advanced, thanks to the help of ML or AI, it should be classified as an AI medicine. That’s the case for all of our early repurposed drugs. We found the use of those molecules that already existed.
What’s different about 3964 and C. diff is, we identified using our platform the use of a specific mechanism in the context of the disease and a specific molecular starting point. Then we built a molecule and optimized the molecule with an in-house chemistry team and using some of our very early digital chemistry tools. So, this is the first molecule in the clinic from Recursion where we not only identified the disease to go after, but we also identified a brand new molecule where we have our own composition of matter [patent] that we filed.
You know, I don’t think patients care who filed the composition of matter [patent]. They care if the medicine works. But for many in the field, that was an important point for them. They wanted to see that not only could we find biology, but that we could do chemistry. And I think ‘3964 has a good shot at showing the world we can do both.
GEN Edge: How does Recursion define its AI-driven approach to drug development?
Gibson: Today, we’re deploying machine learning and AI and many other technology tools across really almost every step from identifying unmet need in patients where we’re using LLMs and other tools to do that work in an automated way, to identifying novel targets, which we’re doing with our foundational approach, with computer vision and machine learning based on images—microscopy images of human cells.
We now have generative AI tools in our chemistry workflows. We’ve got generative AI tools and other ML and AI tools in our ADME and Tox portfolio. More than any other company in the [AI drug development] space, I believe, we have built the full stack where the philosophy of leveraging the right technology at each step has been integrated.
This is the great challenge of our field: There are hundreds of companies today building ML and AI tools. And most of those companies are focused on one part of the drug discovery process. One of the challenges is that you can’t build everything. And the reality is that to truly bring value in many cases to the industry, you can’t just optimize one of the hundred steps to discovering and developing a medicine. You have to optimize entire pathways of discovery and development.
We’ve been criticized for a high burn [rate of capital spending]. But I think a lot of that is because we’re trying to put all of those pieces together because of the compounding efficiencies that one gets as you layer these tools on top of each other. And I think time will tell—the next 18 months with potentially seven to 10 readouts, depending on whether the [Exscientia] business combination is approved and goes through. You know, I think it’ll be an important time for the world to start assessing our rate of success, the time and cost for us to get medicines not only into the clinic, but into early clinical results, that kind of mid-stage of development.
We feel confident about the approach we’re taking. And we feel like we’re going to set a good baseline with these first seven to 10 medicines. And that by the time we get the next 10, and then the next 10, and then the next 10, we’re going to be able to start showing consistent improvement in not only the probability of success, but the time and cost it takes for us to bring each new medicine into the clinic.
GEN Edge: You talked about developing 100 drugs in roughly a decade first in 2014, and again recently, although lately you said it may take 11 or more years. But Recurison’s pipeline only discloses seven candidates plus “more than a dozen” additional early discovery and research programs. How do you plan to build out the rest of that pipeline? Will that be mainly internally or through acquisitions like Exscientia?
Gibson: I think it’ll mostly be internally. Beyond the programs that we have internally, there are dozens of programs that we have not talked about yet, that we will begin talking about in the coming quarters and years. I think our pipeline will have 100 programs in it in the not-too-distant future. Some of those programs will be fully internally developed. Many will likely be with some of our pharma partners.
So, for example, we have a program that’s been optioned by Roche [and its] Genentech [subsidiary] that is advancing in oncology. Should that program continue to advance, at some point we would start to put that into our pipeline and count that as well. Look, I don’t know if it’s going to be 11 years or 14 years or whatever. But we’ve set an ambitious target that forced our team to think very differently about how one might discover and develop medicines.
And if we had 100 programs in our pipeline in 2027, let’s say, and started reading out multiple Phase II programs that were efficacious, I think I would be really happy, even if a little bit late on that promise.
GEN Edge: You mentioned the Roche collaboration, an up-to-$12 billion collaboration that would account for more than half of the $20 billion in future milestone payments before royalties that Recursion has projected could be generated by it and Exscientia if the companies were to combine. Bayer accounts for another $1.5 billion in potential future milestones, plus another $1.3 billion from Bristol Myers Squibb (BMS). What other collaborations account for that up-to-$20 billion?
Gibson: [They include] the Merck KGaA collaboration on the Exscientia side [oncology and immunology, $20 million upfront, undisclosed discovery, development, regulatory, and sales-based milestone payments]. Exscientia also has a really substantial collaboration [oncology and immunology/inflammation, up to $5.2 billion, launched in 2022] with Sanofi.
What’s most exciting is, those are all future payments. What you can see with Roche/Genentech, we’ve had a program option in the GI oncology space. You saw a few weeks ago where Roche optioned the first neuroscience phenomap, triggering a $30 million option fee that came across the line. You see us executing across both programs and kind of these infrastructure building opportunities that will underlie future programs.
On the Sanofi side, there’ve been a number of programs optioned by Sanofi for Exscientia, a number of programs optioned from other companies that have made it into the clinic from Exscientia, some that they have ownership interest in and some that they don’t anymore.
There’s a popular narrative that ML and AI haven’t really done anything in the [drug development] space. And I think that that narrative is already being challenged. And over the next 18 months, we hope to challenge it significantly more.
