Unmet clinical needs often serve as the fundamental driving force behind drug discovery programs. During lead discovery, an intensive search ensues to find a drug-like small molecule or biological therapeutic, initially a hit that, through further validation and evolution, can become a development candidate that will progress into preclinical and, if successful, into clinical development and ultimately be a marketed medicine.
Many drug discovery and development recipes start the same way—with a reliable and scalable assay focused on a specific therapeutic target to search sizable chemical libraries for compounds that have the desired effects on the target. But the time to clinic, expense, and a consistent late-stage failure rate due to ineffectiveness or severe toxicity makes this formula often fall flat, and a primary driver may be that the initial target hypothesis is incorrect or incomplete.
Headquartered in Salt Lake City, Recursion has developed an integrated operating system combining proprietary in-house data generation and advanced computational tools to generate novel insights to initiate or accelerate therapeutic programs. The company uses a proprietary collection of highly relatable, high-dimensional biological and chemical datasets built in their own laboratories and spanning multiple data modalities. Recursion iterates on this approach to create maps of biology that it believes contain more robust and systematic relationships that help them identify one or more targets or hit compounds which may not have been obvious to choose based on public data or our current limited understanding of a disease. This closed-loop system enables Recursion to reimagine the drug discovery funnel, broadening the funnel with potential therapeutic starting points beyond hypothesized and human-biased targets and narrowing the funnel by identifying failures earlier in the research cycle when they are relatively inexpensive.
GEN Edge interviewed co-founder and CEO Chris Gibson to get the scoop on the story behind launching Recursion and how the clinical-stage biotechnology company is thriving in Utah—a growing biotech hub.
GEN Edge: Why did you start Recursion?
Gibson: I was an MD/PhD student at the University of Utah. It’s a long story of how I got there, but I’m glad I did because I ended up joining the lab of Dean Li. He’s my co-founder and was my dissertation advisor. He’s now also the president of Merck Research Labs (MRL). As a trainee in his lab, I was focused on engineering solutions using mathematics and computational sciences in the space of biology.
One of the things his lab had been working on before I joined was a specific disease called cerebral cavernous malformation (CCM). They were leading the way in understanding the molecular mechanisms of that disease. We got to the point where we were confident enough in a potential mechanism—the activation of the protein RhoA—that we built an animal model, which is now one of the gold standard animal models for that disease. We tested a molecule that inhibited RhoA and, to our surprise, it actually worsened the animals, which was a humbling moment, but hints at the complexity of biology.
This failure spurred the idea for a new type of company. How could we take a less biased approach? Could we use images of human cells and computer vision approaches to turn those images into sophisticated omics data unveiling broad, deep and complex biological systems? Would that allow us to take a less biased approach where rather than begin with a hypothesis, often generated with a paucity of data, to an approach where we let the cell, and all of the complex biological systems within it, holistically inform a target in an unbiased way?
Turning that vision into reality wasn’t easy; we essentially created a phenotypic screen, and I started coding up machine learning classifiers. However, I gave up about three days later because I’m not great at coding and it is a sophisticated problem! Lucky for us, we found the work of a woman named Anne Carpenter of the Broad Institute, who’s one of our scientific advisors and a leader in this field. She had built and popularized this new kind of omics called phenomics, based on cellular images in the academic setting. We took that work and evolved and scaled it. Today, we have knocked out every gene in the human genome with multiple CRISPR guides in multiple human cell types and profiled hundreds of thousands of molecules to build maps of biology that inform a deeper understanding of biology’s potential complexity.
Using only images of cells and our algorithms, we can infer the potential relationships between any pairwise set of genes, the potential mechanism of action of a new chemical entity (NCE), or whether two compounds have the same biological effect and function. All of this is possible because in biology, structure suits function; morphology-based fingerprints in the billions of images of cells that we have acquired tell us about the holistic, functional similarity (or not) of any of these queries. As of our last earnings report, we have 2.4 trillion of these relationships mapped out. We’re a company that likes to grow fast and think broadly.
GEN Edge: What does the typical drug discovery journey look like at Recursion?
Gibson: Our scientists don’t need to find a Nature paper and then build a team and assay and go to the lab to start a program. Instead, they can open an app and type in the name of a gene or a compound of interest (or many genes or compounds) based on some anchoring point of biology where we have reason to believe that novel relationships might be informative or essential.
For example, we can query an undruggable target and ask if there are other genes that might encode a protein proximal or distal to that target. Or in the case of genetic disease or genetically driven cancers, we can ask if there are other genes that might encode proteins that are part of the pathway of pathophysiology, but which haven’t been discovered or published yet. We can identify novel relationships like this quickly—in minutes if you do the right searches. And because it is simply a search using an app, and all the experiments to generate those relationships have already been done, our teams can generate new program ideas really quickly. During a Hack Week last year, about a dozen teams generated over 100 new program ideas in about a week at Recursion using these maps.
Once we have a new program idea, we can validate its potential by ordering additional omics assays—we essentially ask if this novel relationship holds up to lots of exploration. While we started in this space of mapping and navigating biology using images, we’ve now scaled sequencing data too. We sequence about 13,000 exomes a week at Recursion. And we use those data as a tool to validate these image-based hypotheses. If you have some drug that rescues some genetic disease model in an image-based screen and also rescues some interesting gene expression profile, you have a lot of confidence that this is worth driving to explore and understand better, at least in a cellular context. We’re trying to industrialize that approach.
And only then, after the orthogonal validation of a variety of omics assays—typically at the lead optimization stage—does our team spend the time to really dig in deeply and begin to build out disease specific assays and deep understanding. We have the freedom to move fast, explore quickly and broadly, and then work on the novel relationships that we can’t disprove.
Our company philosophy is to build technology at every step to decouple a person sitting in the lab pipetting back and forth from the advancement of a program. My analogy is that our automated lab does the equivalent of my entire PhD’s worth of data generation every 15–18 minutes. Our team gets to think about the results of a massive scale of data—now 16 petabytes—and which novel relationship we should translate into a program and drive towards the clinic from that starting point as opposed to the literature alone. It’s just a different philosophy, turning biology into a search problem: can we build a platform that is the Google Search of biological relationships so people can freely explore biology and generate programs that way?
GEN Edge: What has the funding journey been like for Recursion?
Gibson: We founded Recursion in November 2013, and we spent the first three years mostly with angels, friends and family money, and small business innovative research (SBIR) grants from the NIH. We were unlike a Silicon Valley company with a $100 million Series A. It was bootstrapped: a few hundred thousand dollars the first year, $1 million the second year, trying to understand and test this hypothesis about whether you could use images to map biology before we scaled the company.
In 2016, we did some pilot work internally and with some pharma companies, and we convinced ourselves at that point that this was a crazy idea, but worth deeply investing in. In the fall of 2016, we raised our first real venture money—a $15-million Series A from Lux Capital and other investors. We were only about ten people at the time. All of the scaling of the company—we are now about 500 people—has come since.
But it hasn’t all been smooth sailing: there have been lots of little pieces that haven’t worked. For example, our original disease modeling approach relied on siRNA technology. We started hitting our heads against the wall because we couldn’t reliably map the whole genome—only about 10% genes could be well-explored in our hands using siRNA because of all the noise that tool generates. In 2019, based on early data, we made the decision to pivot all our disease modeling to CRISPR at scale, which required a massive effort across the company. That was game-changing.
There have been big technical breakthroughs like this along the way, but all still are generally supporting our original vision to deploy technology like computer vision and ML/AI to understand biology more deeply. And now that we’ve grown, with multiple clinical programs, we have added to our exploratory origins with deep focus on execution and delivery and proving it to our colleagues and our partners with data.
The currency of biopharma is assets in the clinic. Because we have focused not only on our platform, but also on advancing assets into the clinic, we are now taken more seriously— there is a set of proof points that help elevate the conversation, so people are willing to take the time to dig in and understand all of our technology and our unique approach. And that has led to big successes. For example, last December we announced a ten-year collaboration with Roche and Genentech to go after key areas of neuroscience and one specific oncology indication. We signed up together to go after up to 40 programs with novel targets. That’s exciting stuff!
GEN Edge: What is Recursion’s business strategy for generating revenue streams?
Gibson: The most valuable thing we’re building as a company is not our specific programs. It’s the dataset and platform. When a program dies, all that data becomes part of our attempt to do one of two things: either fail less, or fail as early as possible when it’s the least expensive. And that means the next set of specific programs may be better than the first. It’s our hope to continue this virtuous cycle of learning and iteration—in fact we designed our technology to take advantage of that learning, and I hope we can always say that is true.
If you look at Netflix, Tesla, Amazon, there is a story for us to learn in biotech: these companies created a virtuous cycle of learning and iteration. They started with a small dataset, and then they created a mechanism by which real actions led to expanding the dataset. With Netflix, what you click on and what satisfies you, the customer, grows the dataset and informs the algorithm to predict the next suggestions better. That’s what we’re building, but in biology. We’re building our pipeline of programs to generate more and more data across more and more of these layers to make our algorithms better at every step.
We’ve got partnerships trying to make drugs, as well as our own pipeline of drugs. We hope those become medicines that are valuable to patients and create revenue for us, but ultimately, the real value is in improving our platform. In 10 years, if we can make drugs twice as fast, and perhaps at a lower cost than today, that would be great. And ten years after that, we hope the same thing could be true. Can we reach the point where we can make better, more personalized drugs for patients that are less expensive for the healthcare system? There’s a lot of investment and mapping because biology is ultra-complex. But it’s not fundamentally unsolvable.
GEN Edge: What is Recursion’s strategy for growth?
Gibson: The biggest technical hurdle for us to overcome was understanding biological relationships at scale. That’s where we’ve invested much of the last nine years—taking hundreds or thousands of potential programs and industrializing the validation and translation of those to clinical assets. That investment for us looks like things like sequencing at scale, proteomics, predictive ADMET, and more.
We also look at innovations happening outside Recursion that would augment or accelerate our mission to decode biology. For example, we bought a company two years ago called Vium. They put cameras in the cages of mice and use machine learning to identify phenotypes in mice from videos taken in real-time. Does the mouse move differently if it has a genetic disease? Does it breathe differently? Does it wake up at different times? We measure hundreds of things using machine learning on these videos. You can get much more sensitive disease models and identify treatment effects.
Even on the pharmacokinetics side, we run studies in these cages to help us recognize worrisome signs that I think are otherwise missed at that early stage in our industry using more obtuse measures like weight and death alone. The increased sensitivity of our approach means we can raise flags on molecules months earlier. That may not change the game applied to one program, but if you can accelerate every program by three months here and there, when you’re trying to build an army of programs, maybe one day hundreds of programs, saving three months on all of them or identifying the failure that much earlier can be hugely impactful.
Much of our investment is building technology into all those other steps: predictive ADMET, digital chemistry, and automation. Combine all those steps, and maybe we can go from insight to a reasonable, exciting clinical candidate and get it to the regulators for their evaluation in much shorter periods and at a much higher scale. That’s where our focus is.
Down the road, five years from now, I think we want to start thinking about how you connect this more directly to patients. How do you get data directly from patients? Could patients who donate samples or whose samples you use be the patients who become part of your trial? We aspire to get that cycle time down from five-to-seven years to one year. Maybe a patient comes in, and then they’re part of your trial 15 or so months later. Then, you could make medicines for subsets of different genetic or oncologic diseases with decreased cost. Those are the big investment areas for us over the next ten years.
GEN Edge: What are the pros and cons of being based in Utah?
Gibson: It has been a challenge to recruit. But every shadow has an opportunity, and our retention rates are very high. We end up getting folks who are missionaries, not mercenaries, because they’re not worried about vesting at that one-year cliff and hitting their next job. A set of folks do that, and that’s fine—when they go to a startup next door, it actually helps to create a more sustainable ecosystem here. But for us, we’re building something complex and we need people willing to come and stay and invest years of their effort. So, we have spent more time recruiting, but I think we get people who are extraordinarily committed to the mission, which is the upside.
And while starting here had its challenges, as any place does, I’ve been taking many calls over the last couple of years from folks in the Bay Area, Cambridge, and other places thinking of opening second offices or moving here. Denali Therapeutics, for example, is opening an office here. Ultragenyx just opened an office here. It’s enough of an exciting scene on the biotech side that you can make it work. And on the tech side, there’s a lot here. A third of our company is software engineers and data scientists. Many drug discovery companies talk about using ML and AI, but they have 50 biologists and chemists and one data scientist. We’re roughly a third data scientists/software engineers/automation engineers, and then roughly a third biologists and chemists. It requires both of those groups to build. And we can leverage a lot of that tech talent here.