Building a Platform Biotech and The Future of R&D: In Conversation With Chris Gibson

I sat down with Chris Gibson, co-founder and CEO of Recursion, for part two of a series following up on a few pieces I wrote on pharma software and data including Real World Evidence and R&D tools. Chris and Recursion are at the forefront of using AI for drug discovery. We discussed how Recursion predicted drugs that would be effective for COVID, his take on different techbio business models, and what pharma R&D will look like in 25 years. Check it out on Apple or Spotify.

Chris started Recursion in 2013 with algorithms to identify changes imperceptible to the human eye in cells. Since then, the company has built out a set of biological processes and algorithms, strengthened by the data they generate, to broaden the drug discovery funnel at the top and better predict performance down the line. Recursion IPOed in April 2021 and works across a broad swath of disease areas today, both bringing their own compounds through trials as well as partnering with big pharma companies in other therapeutic areas. Chris and I discussed how he’s thought about taking compounds through trials versus partnering with pharma and how the extent of the economic value captured by assets in the clinic has driven Recursion to move toward more of a full stack biopharma company.  

Check out a full transcript of our conversation below: 

---

Chris, to start off, I would love to hear a bit about the founding story of Recursion and how you came to start the company.

I was working in the lab of a guy named Dean Li at the University of Utah on my MD/PHD at the time. I had no plans to drop out and wanted to be a cardiothoracic surgeon. We were trying to understand a variety of genetic diseases and we ended up working on one of those for a long time. I was part of this project to understand a disease called cerebral cavernous malformation, which is a mouthful, or CCM for short. After a decade in Dean's lab and me working on the project for a year or two, we thought we figured it out and we went to test our hypothesis and we failed. We actually made an animal model of the disease worse. So it was like this humbling moment of biology where we learned that we didn't understand it at all, which is not surprising.

And it was in that moment that we decided to take a less, I would say, biased approach, a less rational approach, and actually just to ask the human cells to tell us what the answer might be. So we started writing software, ended up transitioning to software that was already written by a woman named Anne Carpenter at the Broad called CellProfiler that allowed us to basically train machine learning algorithms to recognize the differences in human cells with and without this disease using microscopy images and machine learning models. Then we could add thousands of compounds and identify one that maybe would make those cells look healthy again in an unbiased way. And one of the drugs we identified using that approach is now in a phase two trial for that disease for CCM. That was the origin of the company. I decided to finish my PHD, to take a leave of absence from med school, which I never went back to finish.

When did you know this was going to be a company?

Not immediately. As we did the project on CCM, we wondered, “If we could do this again, could we do this for another genetic disease or maybe 2 or 10 or 100 or 1000?” It was when we started having those conversations that Dean and I decided I would go to Stanford for the summer to a program called Ignite, which is like a mini MBA for scientists, mostly Stanford students, but a few from the outside. I did that; it was amazing. I came back from that and I was like, "Dean, we're going to start a company." And he was like, "No, you need to finish your MD." And then I was like, "No, no, no, we're really going to start a company." He's like, "Okay, well if you're going to start a company then I'm doing it with you." So that was history.

I think it would be helpful for our listeners, before we dive into more of the specifics, to understand how all the work Recursion does fits together. Maybe if we take an example of a compound that you're taking through trials today just to think about how the unique approach you have on the R&D side leads to insights and then a compound. 

What we do is pretty complex. I'm going to start with an analogy. We're in the middle of a vast wilderness and we're trying to figure out how to get to some mountaintop. This is the vast wilderness of biology, super complex, super hard to understand. Today, the molecular and cellular biology tools that most of the industry uses allow us to look at just tiny sections of that wilderness, so we can understand little parts at a time. Ultimately if you want to build a path to some mountaintop, maybe some new treatment for a disease, navigating that wilderness without a map is really hard. And the tools we have today, like I said, only let us see little pieces at a time. So what Recursion did is ask, “Is there some way that we can build maps of this wilderness, maps of biology, that allow us to plot a path from one point to another and to do that at scale?”

The way we've done that is by generating very large omics datasets. So complex, high-dimensional datasets. Our foundation is using microscopy images of human cells, but other omics datasets, transcriptomics as well, that we're scaling here at Recursion. Then we train machine learning systems to try and understand the similarities and differences, not between one disease and another, but across the entire genome and across more than a million compounds. To start to understand the topology of biology, chemistry, and how they all interact.

So how that plays in a single program is we look at our map of biology, we look at some known anchor point like a genetic disease. So we know that patients who have a mutation in a gene called CCM1, CCM2, or CCM3 get this CCM disease. So we can look at our map and ask, “Is there anything in our map that tells us that there's some relevant unknown biology or chemistry that seems to be impinging on this process?”

And we don't have to go do any experiments today to do that. We've already knocked out every gene in the genome and several cell types. We've profiled now more than a million molecules at multiple doses. We do this in a web app where our scientists start with a web app of now over nearly 4 trillion relationships that we've predicted from nearly 200 million experiments we have done in the past at Recursion in a giant automated lab. And we use those to generate hypotheses and then we go test those in our laboratories in a variety of relatively high throughput ways, all the way down through animal models where we've built machine learning systems to watch animals in their terrariums and identify whether drugs could be having not only beneficial effects but also negative effects, like toxicity that we don't want to take into human trials.

It's technology for every step, big data for every step, machine learning wherever it's useful to try and help us make sense of this massive data set. Mostly what we focus on is unexpected relationships in biology and looking at biology as a system. 

There are so many algorithms you're using that are driving decision making. How did you think about some of the early demonstration points to give both you, the biotech community and investors the confidence that you were onto something? 

There are two flavors that we like to look at. One was we asked, “Can we rediscover known biology?” I'll give you a great example if people want to go look this up, so they know that I'm being honest here. I think it was the third week of April of 2020. We published a paper in the first few weeks of the pandemic where we had taken live SARS-CoV-2 virus and added it to primary human cells. And then I think we looked at about 17 or 1800 FDA-approved drugs. And we used this mapping and navigating approach and our technology to essentially predict whether drugs could have some useful effect. So you could go back and look at the pre-print with the original date on it of April 2020. And what we could have told you then was that we don't think hydroxychloroquine is doing anything useful in the context of this disease.

And this is back when that was a hot topic of discussion. Remdesivir, which had not yet read out of a clinical trial or a Gilead drug, looked extraordinarily powerful. And so far today we're about eight for nine of molecules that were in our screen that have gone through randomized controlled trials, have read out trials that align with the prediction we would've made about that molecule. So there's a great piece of leading evidence that the platform we're building is good and that we can do it quickly because we generated that data, we made it all open source, 300,000 images all in the first few weeks of the pandemic. And we did it all at a biosafety level three facility that's like 70 miles away. It wasn't even in our main facility because we're not authorized to work on the virus here. That's a great example, and we have many other examples like that. None quite as dramatic, but many other examples.

We basically rediscover known biology or predict biology that comes to pass.

What's the second demonstration point that you think about?

We basically make predictions using our map and then we go validate them. We've got a really cool program in oncology in an area of ovarian cancer where we made this really novel prediction in our map about a totally novel target. We then took that target and went and explored it in a gold standard animal model of the disease and had a really strong result. When we give statistics about how often we find something and then validate it, that also gives investors and partners some confidence that this is worth investing in.

One thing I'm struck by is as the company has operated longer, you've taken some of these potential candidates into later and later stages of development. And as you’ve done this the capabilities of Recursion have continued to increase (e.g., you mentioned being able to track animal behavior). In a letter you wrote to shareholders you talked about moving from this brute force research to more of a mapping focus in the past year. I would love to hear a bit about that evolution and the broader movement from your core microscopy models to everything you’ve built on top. 

I think the reality is when you discover and develop a medicine, there are hundreds of critical steps along the way. I think many companies like ours have built really interesting technological or process innovations at a small number of those steps, but ultimately you've got to string them all together to get a drug to patients. What we've done at Recursion, we've been fortunate to have really supportive investors and partners who've encouraged us to be bold, to believe that one could build a company that's digitally native and that applies the best technology. Sometimes machine learning or AI, sometimes not, sometimes it's just good old fashioned innovative processes or robotics. But we try to apply innovative steps, not at one point in the process but at many. That's the philosophy we've taken and we've had the supporters that have allowed us to have the resources to do that because I think each of these steps unlocks some value over time.

And if you can stack them together and eventually build a company… and we've got many more innovations to build in the future. We're nowhere near full stack biopharma soup to nuts, but we've built more innovations at more steps of the process than almost any other company in this space. And that's allowing us to get essentially a magnification or multiplication of these smaller effects at many steps as we build that vertical.

It seems like as you go further and further you look at every process of drug development and there are some things you can optimize. I imagine you look at other parts and you might say, "Hey, this actually runs pretty well, we can do this elsewhere." The trial side is particularly interesting. I'm curious how you think about, “Hey, these are things Recursion needs to own and we'll be best in class and we'll be better than how other folks do this,” and other parts where you say “Hey, we don't need to reinvent the wheel, someone else might be better served doing this in partnership with us.”

I think you've nailed exactly how we think about it. We ask the question, “Do we have an insight or an innovation or a technology or a team or a process that would allow us to do something here that we don't think others are doing?” Or in some cases we see that there are others doing really interesting things in a space, but we don't see an obvious way to catch up with them. So that animal technology, we did not build that internally at Recursion. We actually bought a company that we really admired that had made some decisions from a business model perspective that put them in a position where they were not well positioned to take advantage of their technology. We were, so we actually acquired them and then built that into our process.

We want the best process, the best technology, the best team, no matter where it comes from. Sometimes we can build it in house, sometimes it already exists externally and we can just use the software somebody else has built. We love that — if somebody's already built the software for us and we can just subscribe. But it's wherever the best technology is. How do we integrate that into what we're doing?

In your annual shareholder letter you talked about how there's this dual path for your existing programs. You can take them all the way and be an end-to-end therapeutics company or you've done some interesting partnerships with big pharma companies. I'm curious how you've thought about which disease areas or programs make sense to partner versus go all the way yourself.

I think eventually we want to be able to go all the way ourselves in all of the therapeutic areas, but obviously we can't do everything all at once and there's a tremendous amount to learn. We asked the question, “Are there areas of biology where we could be capital efficient?” Rare genetic diseases is a great example. If you look at the clinical trials we're running now for multiple of these rare genetic diseases or areas of precision oncology, relatively small patient populations, there’s not a lot of competition for trial sites. So we're positioned where maybe these things cost 10 or 20 million dollars to run a phase two trial. Contrast that with cardiovascular metabolism, or you're going to go run a weight loss trial or something, or a heart disease trial. You're talking about 10,000 patients and a 100 million plus dollars. That is not a place for us as a company that wants to build a portfolio of programs for us to focus on initially, but it could be a great place to partner because the only companies that can run a trial in that space are scaled companies with extraordinary resources.

So maybe we partner there. And if you look at our partnerships, we have two large partnerships: one with Bayer in cystic fibrosis, a big intractable area of biology, and another with Roche and Genentech in mainly neuroscience, one oncology indication as well. Again, neuroscience — another huge area of unmet need, but where the timelines, the resources required to go into clinic, are really large. So we're partnered with the best in those areas and it allows us to both create value from our platform into those areas potentially, but also to learn from great teams who know a lot about a space we haven't built in yet, so that maybe one day we could build our own clinical development team in neuroscience. But not until we're done with that collaboration and hopefully we get a lot of drugs across the line with our partners there along the way.

You mentioned the resources to be able to conduct some of these longer study timelines, capabilities around actually doing recruitment in these areas where there are tons of trials. Are these partnerships mostly around the trials side or is there R&D collaboration as well?

Absolutely, there's R&D collaboration across the entire space. I think we both learn from our partners about lots of elements of discovering and developing medicines, but also in many cases I think they learn from us and our take on discovering and developing medicines. It creates a really nice opportunity for both groups to be pulled in a direction towards what we hope is a more efficient, innovative, exciting way to discover medicines. And we learned a lot. We hope they learned a lot and so far I think those partnerships are going really well.

As you think about the goal of eventually being end to end, does that mean that in that future state Recursion has got some special sauce on the trials side? It seems like the next one for you to continue pushing the envelope on.

It could be. What I could say is that we only want to build into spaces either we think we have some advantage in building or potentially buying into a space or where we have to. We only want to build into later stage clinical development, commercialization, marketing, etc., if we thought we had some advantage, some perspective that was different or if we had to. I think you see us hedged — we've got some partnerships, we have our own pipeline, we're hedged to try and understand where the industry is going. Is this industry going to want to see us like visa, where they're going to start believing in what we do at scale? And they say, “You know what? We'll run the trials, you guys just do all the discovery across all these different diseases.” Maybe then we don't have to build a much, much bigger clinical development team.

Or what we've learned so far is that in this industry, the currency is assets in the clinic. Regardless of how cool the technology is, people want to see assets in the clinic and proof of concept readouts. And that's where the value inflection point is not only for companies but for patients. That's where people start getting excited. If that doesn't shift in some marked way, you'll probably see us move more towards the vertically integrated side. I think that's how we've hedged ourselves for the last several years, is to be ready for either of those eventualities.

It seems like the evolution of these financial arrangements of handing off a compound at a certain stage is still very much influx. To your point, I think today the pure prize is just getting something all the way through, and it's unclear if you can hand off something before then and still get any percent of the economic value you'd get running it all the way through. It makes a ton of sense that folks are eagerly watching whether that might change. You guys get a few more proof points and it's like, oh man, a Recursion asset at an earlier stage, that is something that we may put a premium on relative to something else.

That's exactly right.

The thing I'm impressed by when looking at your pipeline is the sheer breadth of diseases you're tackling. How have you thought about what the right areas are to apply the platform to? I'm sure early on you were like, “Do we just go after a few or do we do all of these?” 

Well, from the very early days we've been focused on scale. So even in the first few years of the company we were exploring hundreds or thousands of genes associated with genetic diseases. More recently, we've now explored the entire genome. What that means is that our scientists don't need to build a team and build an assay to start a program. Now, to advance a program all the way through the clinic, of course you have to build a team, you have to build assays, etc. A lot of local expertise. But our scientists have the freedom to explore. If you and I had a conversation right now about kidney disease, I could open up our map (it's a web app) and I could start querying kidney disease genes and ask, “Is there some novel insight in our platform?” And maybe there is, and I can just order an experiment to confirm or validate at some high-dimensional way.

What that means is we've lowered the activation energy for a scientific team to get excited about a novel idea. It's now tens of thousands of dollars for us to get to a confirmed and validated novel relationship in biology or chemistry. Now, to act on that means we do need to make a strategic decision. We are mainly focused in oncology and genetic disease right now at Recursion. Those are areas where our platform, we know it works better than other areas. They're areas where we think that there's an economic model that makes sense for us right now, especially in the current capital market kind of crisis in biotech, I would say. So we certainly do narrow the focus, but we're building this reservoir, so to speak, of programs and ideas that can be confirmed and validated, that maybe in two years we come back to this one and say, “It's not right right now, but…” Infectious disease — not a great commercial model right now, but maybe we end up partnering with the Gates Foundation or the federal government to build some interesting extension of our platform in a couple years.

We'd already have some of those insights generated in our maps today. I love that about what we do. It's turning biology from an artisanal bespoke process and drug discovery from an artisanal bespoke process with really smart people spending a ton of time asking and answering questions into a search problem. So the same really smart, talented people can spend their time downstream on a higher probability of success already confirmed and validated relationship.

I'm sure it must be hard not to go chase some of these threads that come up when people are looking in other disease areas.

It's hard all the time. I was just interviewing somebody and we were looking through the map together and there was something I wanted to go chase. It takes a lot of discipline to stay focused. What we've tried to do is automate a lot of this because we want our people focused on not taking a program through all of these relatively repetitive processes. We want them focused on ideating the initial ideas for programs. And then once we've confirmed and validated a relationship that's novel and exciting to us, what's the killer experiment? What's the organoid model or the patient cellular model or the animal model that gets us to believe this is real and worth investing more in?

We want to be a company that fails fast. So why do a bunch of studies to try and understand everything if there's one study that gives you a go or no-go decision? It takes a lot of work to be good at that. But I think we're trying to continually improve as a team to be as good as we can possibly be at getting to that killer experiment immediately so that we don't invest time in a program that we think is going to fail or that we don't think is going to fail, but ultimately will.

How did you think about that in the beginning of the company? I imagine there was probably pressure early on to get some insight in some specific disease area as far along as possible. And at the same time you have this platform you're building that you could go chase in a ton of different directions. 

The good news is when I dropped out of med school, I dropped out not only with this idea for the platform but also with this initial drug that we had discovered with another lab collaborating with us. Tony Donato had suggested we look at this molecule, we put it into our system, we saw great results. So we had a hit from the very beginning and we had it working in animal models. What we essentially did is with a couple of gaps here and there, we sort of developed the platform and this early program in parallel and eventually got to the point where we had enough other hits coming in that we could have multiple programs. So since the very early days of the company, we have had our own pipeline and a platform that are parallel foci. We didn't always intend to take them through clinical development ourselves. That became a necessary place to build because people weren't yet valuing the work that we'd done outside of a small subset of folks. But that's how it arose. It's been from the very beginning.

How did you run the company with some folks working on the specific assets and some folks working on the platform side? How'd that all work?

I think you just create really clear prioritization. So this year we have five priorities for the company and of course we're doing things that are not on that list, but those are important priorities and they're listed in order of importance and it's really hard, especially as the company grows. But what we try to do, and I'm sure we fail all the time, but our attempt as an executive team, as a leadership team, is to give clear priorities and context to our team so that they can make really good decisions in whatever specific area they're working in.

One thing I think that's fascinating about the business is the Recursion Operating System you've built. It’s an impressive amount of robotics, automation software, and data infrastructure and you built a lot of it from scratch. There's been this rise of a lot more techbio companies, folks leveraging AI for drug discovery in different ways. As you think about the landscape today and the opportunity, is there a place for folks to maybe take some of the things you've built on the Recursion Operating System, build those as standalone companies and sell that into this set of techbio companies? I ask because it seems like from the outside, every company that starts doing that eventually just becomes a therapeutics company. 

It's a great question. We are starting to see this. There are a whole bunch of companies today — in fact we just were talking to one the other day that's built a really nice system for essentially organizing your chemistry and your chemistry experiments and the workflow of the company. We built our own tool to do that and it's not pretty. It's not something that we would want to sell externally or that somebody would pay us money for, but it works really well for us. We were not a customer for them, but if we were starting over and building Recursion today as a 21st-century therapeutics company, we would immediately go use that tool.

I think there are a bunch of companies like this — many, many companies like this that are rolling up and you see some of the biggest ones, like Benchling and some of these other ones, who are building a lot of the software tools that we actually are customers there. But a lot of companies are building tools that if they weren't there, we'd be building that software ourselves. So we love that they exist and I think for a lot of other techbio companies, it's going to be a great opportunity to shortcut towards eventually what your primary goal is and not get distracted by all the infrastructure along the way. But for us, those things didn't exist when we needed them. So in many cases we just had to build it ourselves.

Do you think those companies will work as standalone companies? The industry today values assets all the way through. Do you get sufficiently rewarded if you're handing off something before then? Take the chemistry software you mentioned, what might you be able to sell it to techbio for versus taking the advantage it provides and trying to become a therapeutics company? Is there a path forward on just doing the software side or is it merely a means to eventually build out something broader?

Absolutely, there's a wide variety of business models that could be successful here. I think it depends on what impact you want to have and how you define success. If generating a multi-billion-dollar business with software is successful, which I think most people would argue it is, then there are going to be tons of opportunities to do that. If you want to discover and develop hundreds of new medicines, building a software company is not going to be. So it depends what your goal is. I think we're also seeing a shift. The current capital markets crunch is making it really hard for a lot of late stage private companies in our space. It's better to be public and later stage or to be in the very early days right now. If you're a series B, C or D therapeutics company, but you don't have clinical assets, it's just a hard time.

So that's going to create this gap. And what we're going to see is probably a new set of business models that arise. Maybe a lot of the newer companies aren't going to end up pursuing individual therapeutics. They are going to pursue this software model and they're going to sell into large pharma biotech companies who are now really starting to understand the importance of thinking in a digitally native way. They're seeing this from their employees, they're seeing this from companies like us who are pushing the field, and so they're ready to start being customers of these companies. So I don't know where the landscape will shift to, but it is clear to me that it is always shifting and you've always got to be hedged. You've always got to be ready to react to those things. You can't control how the world around you changes.

It's an interesting point and a nice segue to reflecting on the AI for drug discovery space as a whole. You were one of the pioneers in this space. I think it’s become even more the topic du jour at every conference these days. I’m curious, taking a step back, how you categorize the space today as a field relative to when you started Recursion in 2013?

It's extraordinary. There are so many incredible people, incredible ideas, incredible technology in this space. There's also a whole bunch of hype. We're at this important inflection point where a subset of these companies are getting to meaningful milestones. We're going to be reading out phase two trials in the next few years. That's a really important milestone, not only for us but for other companies in the space. You look at, Relay has read out some trials, Exscientia has programs moving into the clinic. You see companies that have partnerships where you're starting to see programs move into the clinic with partners. These are huge opportunities for our field and it's super exciting to me that there are so many different companies that are moving in this direction because ultimately, of course I want Recursion to be successful. But at the end of the day, if we're not, I want the field to be successful too. Because I think it has so much promise.

What we've got to guard against though, are of course the subset of folks who are making claims that are difficult to back up and they have the opportunity to tarnish the space generally. There are going to be failures, there are going to be successes, but you need to be upfront about the probability of success with folks and hopefully we'll meet — maybe exceed — but hopefully just meet the probability of success of the rest of the industry to start and we can build from there.

Of all this stuff happening right now, there are so many different approaches. Outside of what Recursion is doing, what do you find most exciting of some of the progress in the space?

I think there's a whole bunch of interesting stuff happening in chemistry right now. A lot of generative AI, obviously looking at what AlphaFold and others have done in the space of protein folding and what that's going to enable for biologics and ligand protein interaction predictions I think is super, super cool. That's a space I'm really actively watching. The other space that I think is further from where we are today but is very interesting to watch is the clinical space. There are a whole bunch of early companies that have tried to innovate in clinical trial design, some more successful than others. Obviously innovating in a highly regulated space is hard to do, but I'm watching those companies, I'm looking for successes and failures that can help guide where we go or who we partner with.

I think especially as large data sets from electronic health records become integrated with large genomic data sets, there's an opportunity for one or more companies either to build into that space or for us to partner into that space to potentially innovate in the way trials are done. We're seeing this in oncology and other areas. You've seen some really cool innovations in the way that trials are done, innovations that I think are meaningful for patients and meaningful for companies as well.

Certainly there seems to be tons of changes around everything from study design and easier recruiting, but also virtual control arms, pragmatic trials; it's a pretty exciting space. 

Or even in our own work, we have cameras in the cages of animal models. And we can predict so much about how an animal is responding to a therapy, both good and bad, simply from a video sensor in their cage. Now we're not building here, I'm just totally talking out 10 years from now, but I believe in some form of wearable or some sort of sensor or monitor system because I see what it can do in another physiologic system. Now, I'm not suggesting we go put cameras in everybody's homes. But at the same time, you know that if you could have the right sensors in an environment, you could gather a lot of data that could be helpful in some context of biology.

So what sensors? Is it the one we wear in our watch? Is it going to be the Alexa system in our house? And Amazon's starting to come in. I saw that they were doing a phase one trial recently or helping with the phase one trial in Seattle. So you see Amazon coming from one side, Alphabet with DeepMind and Isomorphic is coming from one side. The pharma companies are starting to invest in this space. What an exciting time to be working at the intersection of technology and biology.

And I think to your point, rather than just read out one endpoint, the amount you can learn from gathering so much more data in a trial makes a tremendous amount of sense. At the broadest level, as you think about where this whole world is headed, I'll ask the highest-level question: What do you think drug development looks like in 25 years?

25 years? Oh I love that question.

Take your number of years that is in the future.

I love it. Because you have to think along that kind of time scale to see a really massive shift in this industry. But I think in 25 years we'll be building N of 1 medicines at scale for most patients with most diseases. So, most people won't have the same disease as almost anyone else. You'll have a version of that disease that's pretty specific to you based on your genome, your environmental upbringing, etc. You'll have a pretty specific-to-you disease and we'll be able to either identify the right medicine for you or even build a novel medicine for you, give it to you in an economic model that works for humanity and that affects most patients with most of the diseases we work on today.

25 years, it's going to be vastly different. Five years not so much. We overestimate what we can do in 5 to 10 years (I think this is a Bill Gates quote) and we vastly underestimate… or we overestimate what we can do in a year and we underestimate what we can do in 10. So if you're talking about 25, I think it's going to be extraordinarily different.

What does that mean for the pharma business model? I feel like today you've got these blockbuster drugs, like Keytruda, and the reality is they work really well for some patients and don’t work as well for others. They're these broad things that are used across the board. It makes total sense that we'll get to a world in which there's something that's much more tailored to each patient. I wonder if that ends up needing to be priced like rare disease drugs are priced. It seems a very different model than what we have today. I'm curious what that means for pharma companies of the future.

I don't know because we're building something a bit different, but what I could imagine thinking if I were at a larger pharma company is I better be hedging and getting into all of these spaces as much as I can and I better be thinking about different economic models because the reality is the system in the US will not exist in 25 years because it is not sustainable. So how are you positioning yourself to have inexpensive medicines for patients and still build a profitable business that your shareholders can get behind? What that looks like to me is trying to find ways to move failure earlier in the pipeline. So if you could eliminate all phase three failures and move them to day two using an omics system, and of course that's not possible today, but if you could make that shift, you could start to make the discovery and development of medicines much more economic.

And if you can keep pushing failure earlier, build better and better models that allow us to explore biology and chemistry very broadly and say, “This is the drug for this type of patient.” There will still be Keytruda drugs from aspirin to Keytruda that work for many people, but we will have very, very few drugs where we don't know ahead of time which patient it's going to work for. So we're not going to have patients going on Keytruda who end up not responding. That's going to be really, really rare in 25 years. I think we're going to know it's going to work for you, Jacob. It's not going to work for you Chris, but for you Chris, it's this other drug.

And maybe for you, Cindy, none of these drugs are going to work, but here's why. Based on your DNA and everything we know about you, it's probably this and we've already got a molecule or a medicine or even a gene therapy that we think is going to modulate that in a way that makes you resistant to that disease in the future. I think in 25 years that'll be happening pretty broadly.

It's an incredible vision. We always like to end with this quickfire round where we ask you a few quick questions. To kick it off, people talk about AI for drug discovery a lot. What do you think is the most over hyped and under hyped thing in the space today?

The most over hyped thing is the idea that AI is going to do it all and it's a magic bullet that's going to cure everything. The reality is AI is a really useful tool, and it's more useful in some places and less useful in others. There's still a lot that people are going to need to do even in 25 years to help discover and develop medicine. I think that's the biggest piece of hype, is that it's not like this silver bullet.

It doesn't automate away the need for lab scientists.

No, no, no. There's still a ton of that work and there will be for a very long time. The most under hyped thing in our space is people. All of the companies talk about their technology, but a lot of the companies don't spend enough time talking about their people, their culture. And the reality is whether you're a tech company, a biotech company, no matter how good your technology, your AI is, if you don't have great people and great culture, you are not going to be successful. So I think we need to spend more time talking about that as a sub-sector of the industry.

You started Recursion in 2013. If you could go back now with all the knowledge you've accumulated and do it over, what would you do differently?

I would've used CRISPR earlier or invented CRISPR. Because siRNA was a really rough technology that we used in the early days. No, but joking aside, I would've hired some more experienced managers and some more experienced drug hunters earlier into the company. I undervalued both of those things in the first few years of the company. And it's not that they make everything better, but you need the right balance of innovation and useful naivete alongside experience, wisdom and honed talent.

A lot of R&D and pharma as a whole touches the policy world. If we gave you a magic wand and you could change one thing on the policy side, what would you change?

I think that our regulators have a really hard job and they do it pretty well. So there are not a ton of policies that come to mind, but one that I would get rid of — and I'm going to get some flack for this — but I would get rid of direct to consumer advertising in our industry. You watch these TV commercials occasionally and you don't have any idea what the drug does, and I don't know why people are spending money on them. I would move our industry in a direction where we generated data sets and we shared them with physicians and patients openly, wherever they could access them, like on the internet where you can search for your disease, and we let people look at the data and judge for themselves and let the medicine stand on the merits and have a little bit less marketing behind them.

We’re one of the few OECD countries  that actually allows that, though it does seem to help the sports leagues with their advertising revenue.

I'm sure it helps somebody, but I'm not clear who it is and why we're doing it. So hopefully Recursion is never one of those companies that... We'll talk about Recursion, we'll advertise what we do, but I really want to avoid these super generic, “Have you ever been sick? If so, here's a drug that could cure you. And oh, by the way, here's the 600 things that it could do that are bad.”

That part they say very fast.

I just don't get it. Anyway. But that's me. I'm naive, I guess.

This has been a fascinating conversation. I'm sure folks will want to go dig into more of the details here. What's the best way for people to learn more about what you're doing at Recursion?

Follow us on Twitter, follow us on LinkedIn. We're always posting about the work that we do, the people, the technology, the processes, our partnerships. You can look at our website, you can listen to this podcast to hear about us and lots of other cool groups that are working in the space. And I think just following us online is the very best way. Or if you happen to be at a conference or something, come say hello.