Good morning, everyone, and welcome to the Recursion company presentation. My name is Ethan Taylor. I'm a vice president in J.P. Morgan's Healthcare Investment Banking Group, and I will be moderating this session. It is my pleasure to introduce Ben Taylor, CFO, and Lina Nilsson, SVP and Head of Platform. So, Ben and Lina, thank you for joining us.
Thank you.
Yes. So, to begin, maybe we could level set and begin our conversation with an introduction to TechBio. So, Recursion is seen as one of the leaders and pioneers in the TechBio space and really helped define it. So, what is TechBio and what problems does Recursion look to solve?
Yeah, so one of the fundamental problems that the healthcare industry has had is a 95% failure rate in going from a scientific idea to actually developing a drug. And so, if you can think of TechBio as having a core mission to change that probability of success because it underlies a lot of the scientific innovation and the cost that holds back parts of biopharma development. And so, what we at Recursion do is think about what are all the reasons a drug fails as it is being developed? What are all the reasons that we see it fail in clinical trials?
And then, can we create a better predictive model using AI and other technologies to be able to design a better drug, to be able to understand the technology better, to be able to create a better patient selection tool so we get the right patients into clinical trials? And so, we do a lot of different technologies, but the core goal is to bring them all together. We have also improved the time and cost pretty dramatically. We've seen an 80%-90% decrease in most of the costs and times associated in drug discovery, but the core focus is really on changing that probability of success, making a 95% failure rate a lot less than 95%.
Great. So, AI and machine learning is being embedded across the landscape in all industries. Why should it be utilized specifically in healthcare and biotech, and what are some of the tangible benefits you're seeing so far?
Yeah, I mean, I think at the start of that, you know, the need, the why, right? We spend 18% of GDP on healthcare, and that's growing. All of us, me, you, everyone we love and care about, it's a past or future patient. So, I think that sets the stage. It's essentially a data problem, right? Data points and how can you bring that together to drive insights on the tangible benefits we're already seeing? There's more on this in our earnings calls and so on, but just to give some quick stats, right?
We've shown a 50% decrease in the amount of money it takes to get to IND, so the FDA giving clearance to begin clinical trials, months of reduction in how long it takes to get to key stage gates in drug discovery, like having a drug candidate, for example, massive reductions in the number of compounds, the potential drugs you have to iterate on, so from taking it from thousands to hundreds. So, we're beginning to quantify the metrics along the way of what is a, you know, very long and complicated process to get a lifesaving drug on the market. But maybe just to wrap the answer to your question, the thing that I'm personally really excited about is that you can also do things that you fundamentally couldn't do without AI.
And a recent example I really like for people who want to dig in more is last August, we got FDA clearance for RBM39. So, this is a drug target that would have been undiscoverable without the kind of AI technology Recursion has. And that's just super cool, right? A whole new avenues into human health.
Yeah, and I think that's a really important aspect, is we're actually able to do things that traditional methods weren't able to accomplish. And that's because biology is incredibly complex. It's a multivariable problem. And historically, what we've seen in traditional methods is a very sequential, simplified process of trying to solve those problems. And so, what we're able to do is create complex simulation and modeling systems that actually can put thousands of different variables and weigh them independently and then as a total so that we can actually come up with something that's thoughtfully balanced across all of the different risks rather than just optimizing around a single area.
Great. Thank you. What are some of the challenges that you and other companies in the space might encounter when you're trying to combine biology and technology?
That makes me think of, you know, there's a famous quote, "Just because it's simple doesn't mean it's easy." You know, it's simple, like the answer to your question in a sense. It's all the things you think it is, but it's not easy in that the secret sauce is for companies like Recursion to figure out durable ways to address them. So, for example, right, different same words having different meanings, different cultures on how you deliver value. Maybe fundamentally, you know, the pharmaceutical industry kind of has this. In the 1890s, right, lots of really important processes and insights are embedded in that industry. But at the same time, tech is all about innovation and questioning assumptions. And how do you take the best of both of those worlds and bring them together? That's a really big challenge.
Great. Thank you. And you, Ben, you kind of touched on this a little bit, but how would you describe Recursion's platform to someone who's not a biotech expert or very familiar with the clinical development process?
Yeah. So, I really think the heart of it is diving into how are we going to make these complex problems and embrace the complexity rather than trying to simplify it out? Because a lot of the failure in the healthcare system comes because we take a very complex problem and we substitute very simple problems for it, and we pretend that we've solved it. And so, what we are trying to do and how we design our platform is to say, "I'm actually going to allow it to be as complex as it wants." So, with our phenotypic platform, rather than taking a small slice of biology and putting it into a separate environment, we let the cell react as a whole.
We can look at the image of the cell as a whole and understand that biology in totality rather than trying to break out little pieces of it. Or with our chemistry platform, being able to say, "We're going to take something that's effectively an infinite space of potential chemistries, and we're going to quickly learn our way into what is an optimized molecule that's going to solve this series of two dozen problems that I need to solve." And so, those are the places where we're trying to allow it to be complex. We're sort of allowing all of the potentials to be out there, and then we're quickly learning our way into what is the solution then that gets us closest to an ideal property.
Great. So, specifically within the platform, I believe there's, you know, kind of four areas you help accelerate identification of a compound to the clinical stage. And that would be quickly validating our hypothesis, designing the compounds. The other benefit is spending less and reducing time. So, if you can maybe, you know, expand upon some of those four benefits in your path from early discovery all the way up to the edge of going into the clinic.
Sure. I guess start with kind of zoomed out answer to that. I mean, it's a tech conference, so at a 10,000-foot view, I think it's very similar to how SaaS companies kind of took over the world a decade ago, right? Anything that you can do in silico in the digital world is going to be just dramatically faster, cheaper, more scalable than things you do in the physical world, right? It's harder in pharma in some ways because you need those models to be really good. But if you crack that piece of the puzzle, you kind of off to the races to really dramatically change how all the different steps happen, right? Because you can do a lot of the insights in a virtual sense, rather than physically executing experiments. Maybe I'll give one example to make it real.
We have a collaboration with a company called Enamine. They are one of the major makers of chemical compounds or potential drugs in the world, and they had a library called the Enamine REAL. It's tens of billions of compounds. Imagine if Recursion were to say, "Let's buy all of those, have Enamine synthesize all of them, ship them to us, and test them." Even with massive robotics, it's just an absurd thing to try to do, so, instead, we worked with Enamine to run our models across that gigantic space and just pick a couple of tens of thousands of compounds that were the most promising, and then round those in that way that Ben described to kind of gather broad biological insights, so, all of a sudden, because we work in silico, the physical problem becomes tractable and fast.
That's true for all these different pieces that you called out.
It's interesting because it's actually an emergent property that we are so efficient. The goal is to make a better product. The fact that it costs a fraction is really because we are using more modern technologies, we're doing it in different ways, we have a better process. Most of the industry is very experimental-heavy and sort of making a guess, checking a hypothesis, and going back and forth in that way. By being able to create highly predictive simulated environments, we can just cut so much of that time and cost out. The goal is always, "I want to come up with a better product." That's why if you look at our collaboration with Sanofi, for example, we've been able to advance four different programs that weren't able to be solved in more traditional methods.
We did it in a short period of time. We've done those four programs plus some others that we haven't talked as much about within a couple of years. That's something that usually would take a decade or more across these programs and cost tens of millions of dollars. That's how the technology comes in to make two different differences: achieving something better, but also doing it far more efficiently.
Great. Thank you. Could you talk a little bit about how Recursion's platform has evolved over the years? I believe now you're at the Recursion OS 2.0 platform version. So, how have you gone from those early operating platforms to where you are today?
Yeah, sure. So, 2.0 is what we call our newest platform post the merger between Recursion and Exscientia. It's really about two basic pieces. One is a massive increase in the kinds of biology we can do. So, multi-omic assays, different layers looking at not just phenomics, the assay that Ben called out, but transcriptomics, genetics, patient data, toxicities and liabilities, and so on. And then, on the other side, the modeling and compute. You know, when Recursion was founded in 2013, you know, the compute that was available at the GTX chips from NVIDIA, they were like golf carts compared to the Ferraris, right, or the H100s we have now on our supercomputer. It's a totally different world. And then, on the modeling side, you know, today we have transformers, we have specialty models at Recursion and for applications like in pharma.
So, for example, just to give one that we use quite a bit, GFlowNet, it takes kind of the best of reinforcement learning and the best of generative modeling. So, basically, what that means is it lets you combine both exploring the unknown and being efficient at the same time in a tunable way, right? It's exactly what you want to do in pharmaceuticals. So, that's to say today we have these ingredients of molecules in biology and chemistry in the physical world, massive compute capabilities, and not just one model, but many different modeling approaches to get at problems. And the end result of that is that, you know, when the company started, it was kind of we had powerful but limited point solutions, and now we really have an end-to-end system. And so, that's what we call 2.0, this ability to cover the end-to-end.
Great. Thank you. So, turning a little bit now to partnerships. So, Recursion is really a leader in both the scope, scale, and success of your large pharma partnerships to date. Could you talk a little bit about how you're using partnerships as a way to drive value for the Recursion platform?
Yeah. And this comes back a little bit to our mandate as a company. It's a little bit different because we had a lot of investors come to us and say, "I don't like biotech investing," or, "I'm not a biotech investor. I want to see the paradigm change." And rather than it be a binary risk, I want it to be a business model. And so, from the beginning, we have always tried to do two things in parallel. One, advance a really high-value internal pipeline, and that's currently making its way through clinical trials. But then also maintain really strong partnerships. And in fact, a lot of the original funding for the companies came from partnerships rather than investors coming in. And so, what we have done is brought in a little over $450 million from partnerships.
So far, we've got hundreds of millions more per program that we can bring in. To give a sense of the value that we bring into these partnerships for our, you know, if you're Roche programs, for example, we can, per program, so per idea, we can get $335 million worth of milestones plus royalties on the back end. For Sanofi, we've got up to 15 programs we could do that on. For Roche, we've got up to 40. What we're trying to do is deliver on areas where their internal capabilities couldn't historically do it. When we talk about hitting the Sanofi milestones, Roche just paid us a nice $30 million milestone for delivery of a map. We have multiple programs advancing with them too.
We're actually not only funding the company, not only building our platform, but also really advancing a high-value pipeline that we have a lot of financial stake in.
Great. Thank you. And where else do you think you might see additional partnerships moving forward? And, you know, what do you look for in a potential partner?
Yeah. I'll start on that. So, we look for partnerships in many different ways. I think one hand is like we were just talking about with Sanofi or Roche, where we've got these broader pipeline partnerships. We also do business development on our internal pipeline and think about, "Should we be doing co-development? Do we want to out-license some of these programs and have someone else develop it so we don't have to build up the later-stage infrastructure for some of these things?" But in addition, we do a lot on the tech and the just biopharma space in general. So, Lina had brought up Enamine. That is a great way to go deep into chemistry without having to build it all ourselves and spend it all ourselves. Of course, our highest-profile tech partnership was with NVIDIA.
You know, Jensen has been a big believer in what we're doing because, in his own words, he sees healthcare and what we're doing as being one of the highest impact areas for AI. And so, they've been a huge supporter, and we continue to dig deeper with them. But we do a lot of other tech partnerships. I don't know if you want to talk about Tempus or some of the other data partnerships.
Yeah. We also have partnerships around patient data. So, the ability to integrate anonymized large cohort data from patients in order to inform really our whole drug discovery pipeline from the very initiation of what does it make sense to go after, all the way through into how we run our clinical trials. Yeah.
Great, and, you know, that's a big bottleneck in the drug discovery and drug development paradigm is clinical trials, right? The speed at which you can conduct them, the probability of success. So, a little bit about some of those partnerships you just mentioned. How are you using your platform and some of those partnerships to help optimize the clinical trial part of drug development?
Yeah. I could take this one. I mean, one big bucket is real-world evidence, right? So, these are partnerships we have to have this anonymous patient data to be able to inform our patient cohorts, you know, avoid drug-drug interactions, select the right patients for the trial, find the hospital sites that have the right doctors working on the diseases that the trials are going to be on, and so on. The second big bucket is maybe, to me, was less intuitive when we started this work. And this is around trial efficiencies, right? Kind of counterintuitively because clinical trials are so standardized and so regulated, there are massive opportunities in order to bring in technology to automate and streamline work, right? One of the big cost drivers for trials is how long it takes you to enroll patients and just run the trial, right?
So, anything you can take from a week to have that return message to get a relationship signed with a doctor, let's say, right? If you take that to a day, you've saved that week, right? So, there's this huge efficiencies that we are going after on the operational side. And maybe if I can give a third example, this is my favorite, is this, and I alluded to it earlier, this idea of bringing in patient data to the very beginning of the process. So, what we're doing is building machine learning models on top of both our in-house data that Ben mentioned, these gigantic data sets, and patient data together. And what you get then is kind of the benefits of both, the controllability and completeness of your in-house data and an improvement of, like, the notorious problem with patient data is the poor signal-to-noise, right?
It's hard to pick out the important stuff from the messy world that you're in. You always have fewer patients than you want and messier data. But combining it with our in-house data and these models, at the very beginning, we get much better signal-to-noise. We're able to pick out insights that you could never pick out if you only waited till you were at the clinical trial and you only had the patient data. It's one of those emergent kind of properties like Ben talked about that, frankly, kind of blew my mind when we first saw it.
Well, and this is all underlying what is a fundamental misunderstanding I think a lot of people have around clinical trial failure, is people think clinical trials fail because of biology. It's one of the potential causes, but clinical trials fail because of statistics. It's how many of your patients responded versus how many didn't. And so, if you think about that, you can have failures for chemistry. Let's say the chemistry caused a side effect that was unexpected. If that patient leaves because of a side effect, they are counted in the same way as someone who didn't respond to the drug. If someone obviously doesn't have the right biology, then that's a failure. If the clinical trial design isn't good, we do all sorts of simulations where we actually run simulated clinical trials to see what the weak points are.
That's such an important component to be able to create that statistical analysis that's going to be successful. And as Lina was just talking about, the patient enrichment, I mean, think about one of our drugs, all of the other drugs in the class have this particular side effect profile. Hyperbilirubinemia basically means that you're producing too much bilirubin, which can be toxic with your liver. And so, what we were able to do is look at that and say, "Well, actually, these patients are almost always going to have drugs that they take alongside this drug that cause liver injury." And so, if you don't design that out, that drug's going to fail, and you can figure it out now, or it's going to have a really small population. And so, this is how we think more comprehensively from the very beginning. We bring it into the beginning.
We say, "What is this patient going to need at the end? What is their actual clinical experience going to be like? What drugs are they going to be on? What is their profile as a patient?", and try and design it all back into that beginning, and that's, I think, why you've seen a lot of our partners, you've seen some of our early drugs. I mean, we had a drug that just had clinical data come out back in December. We were within about 5% of the core modeling aspects that we were putting for how this drug should be absorbed and metabolized and excreted and how it would hit its target. This is all the critical things that are usually more guesswork in a traditional system.
This is why what we're doing is making such a difference, and we're seeing it in each one of the drugs we do.
Fantastic. Thank you. Just to kind of close out some of the conversation around partnerships and tech, you mentioned the NVIDIA one. How do you see non-pharma companies entering into the biotech industry? What value could they bring, and how do they help bring innovation?
I mean, fundamentally, I think that the company that's going to win is going to be a company like Recursion, right? That's not pure tech or pure pharma. That's why I'm here, right? I think that's the winning recipe. But you're right. I mean, there's a lot of companies coming in from pure tech, new and old, big and small, and they're bringing a lot of creative energy, questioning assumptions in the pharma industry, and that feels really good. We have companies like NVIDIA. And that's not just about the GPUs, right? It's about there's fundamental differences, as you can imagine, about modeling text or internet images and data from medical fields and having more infrastructure from NVIDIA and other tech companies to handle the kind of unique data we have in efficient ways is a really, really big deal to be able to build the right kind of models.
So, that's just really huge. And then maybe a final point to add, I think tech companies have brought to the industry is a lot of transparency, open-source publishing around processes and data sets. And Recursion is part of that community. I'm really proud of that. So, I think we've open-sourced one of the, I think it's still one of the largest biological data sets ever. We've shared basic versions of our models. And this kind of community around process has really accelerated what is possible in the pharmaceutical field using AI. And that's really huge.
Yeah. I also think this is probably a good point to talk about the importance of data because I'll be the economic pragmatist here, sort of suiting to my role. I think what big tech wants is for the entire industry to do business in our way, right? They want it to be high-compute, data-rich, cloud computing for some of them. You know, they want people to work in that way, not in laboratory experimentation with on-prem data storage and sort of each experiment is its own experiment. And so, if that actually converts, it's an absolutely massive market for them in being able to exploit from the tech side. And so, what we've actually seen is even, it's funny, even someone like Google, and they have obviously Isomorphic there. Google is a great partner of ours, right? Like, we have a terrific relationship with them.
They are focused on converting the industry to work in our way. And so, I think that's a really important point where I think we have an advantage over a tech company who wants to expand in space is we have from the beginning understood the importance of proprietary data. And so, for more than a decade, we have been doing what at first is a lot of guesswork. You don't know what data is going to be important when you're trying to decide, "How am I going to build a predictive AI model?" And so, the reason we've got 65 petabytes worth of data is because we first started exploring, and then we said, "Oh, this is really helpful, and now this is really helpful." And now we've done, you know, a decade's worth of projects going into that data. But you have to annotate it right.
You have to collect it right. You have to ask sort of questions that are going to be able to inform in an AI-first system. And that's just very different to the sort of, that is a very biotechy side of thing is understanding data production versus the tech side, which is more on the analysis. And you really need both to be successful at what we do.
Great. Thanks. So, you have a very catalyst-rich calendar coming up over the next, you know, one to two years. What are you most excited about, and how will some of those data readouts, trial initiations help, you know, validate your platform?
Yeah. Maybe I'll start here, and then you can jump in. So, there's a couple of different points of validation we always look at. Early on, I think you do see the partnership validation come in, and that's something that we've talked about before. But what we're really going into over the next 18 months, we've seen a couple of early tastes of it with our CDK7 and our FAP programs having some really nice positive early clinical data. So, we're expecting to see more of that come in on both of those programs. And then we've got three other oncology programs that have very clear endpoints. You can look at it and say, "Is this working or not?" And I think that's critical because there's two levels of validation that come in. One is the validation around the platform. Like, is it doing what we designed it to do?
Like I talked about CDK7 before, a whole bunch of different parameters no one's been able to solve before. Are we seeing it in human biology? Check. We've seen that is happening. Then there's a second level of validation of, "Is this actually a product that's going to be important for patients?" And so, that's where you need the larger data sets. That's what we're starting to turn over the next 18 months as we go into combination data for CDK7, as we get a bigger data set for FAP, we start to see some of those other oncology programs readout. So, I think that second level is something everyone can understand because then you have a product.
You can say, "Okay, I know what my patient population is, I know what my usage is, I can create a model around that." But what we've started to see already is the platform validation that it's doing what we wanted it to do.
Great. So, how is Recursion navigating some of the uncertainty and change at the FDA or the larger macro environment right now? You know, recently, the FDA announced that they're going to start phasing out animal testing for preclinical trials. So, maybe starting with that first, how does that impact Recursion, and how is Recursion positioned to address that plan change?
Sure. I mean, the initial guidance from the FDA is from monoclonal antibodies, which is not kind of our angle into therapeutics. That said, right, you know, I think we share kind of the perspective with the FDA that, you know, animal models sure can be great, but animals are not people, right? Humans don't even have the same number of chromosomes as rats or mice, right? Chocolate is poisonous to dogs, certainly not to me. So, you know, we're building models for modeling human effects of drugs, right? What is the drug likely to do in your liver, in your heart, in your kidneys, right? And we have a whole suite of models for this at Recursion.
And then, in addition, what we talked about a bit earlier in this conversation, these models that are around modeling broad swaths of human biology by combining patient data and our in-house cellular data. So, we're, you know, deeply invested in modeling human biology directly, and that's a core part of our strategy. Yeah. So, that's really aligned with the FDA guidance. Yeah.
Great and then what about some of the larger macro news? You know, I'm often asked the impact of tariffs. No volatility there. How are you ensuring that Recursion stays nimble and is able to, you know, weather whatever market turbulence we might be seeing right now?
Yeah. And this goes back into we try and focus on having a business model rather than a lot of binary risk. So, I think there's a couple of different levels. One, having both the U.S. and U.K. operations can be helpful in managing some of that. We actually don't think that we've got much tariff risk just because of the way that we do business, but we will see where all of that ends up. And I think that what we look at on the macro risk is it's obviously been a rough time in biotech for a long time. And, you know, developing new technologies in that space can be really difficult because people are very focused on, "Give me some single thing that I can make my bet on," because there's so much risk associated.
What we've been able to do is sort of balance our partnerships along with doing the pipeline development, being able to address different investor bases as well, because I think there are a lot of people who want to see change in this industry, and that's a really important part that comes into it. We've also seen that from the regulators, as Lina was just talking about with the FDA, want to change some of the things in the system and move more towards this sort of way that we do business and believe business should be done. So, I think we're pretty well positioned. There's a bigger, deeper point that I talk about.
The reason that biotech and pharma in general has been underwater for such a long time is because if you look at not only the risk profile of developing a drug, I mean, it takes close to 15 years, close to $2 billion, and, you know, the drugs don't always perform like you'd want them to in the market. A lot of that is exactly what we're trying to address, right? So, if you think about the pharmacoeconomics of the system, so why do drugs cost so much? It's because there's such a high cost to being able to develop them. Why is access such a problem? Same thing. If you go back and you actually change the inputs and you say, "Well, actually, I can develop a drug much faster, much cheaper.
I can get to a higher probability of success. You address a lot of those fundamental underlying macro conditions, and you really put biopharma development, drug development back into a place where it can be an economically viable model.
Great. And to close out, where do you see Recursion in five to ten years from now, from both a capability and business model perspective?
In five or 10 years, we're going to have that flywheel fully working that we were talking about, data-informing models, models improving what data, narrowing and improving what data we even have to collect, right? Those efficiencies that we're talking about are going to be really there in kind of full scale, right? And then maybe on a personal level, I hope in five to 10 years, I'm getting some hugs from patients who took Recursion medicines.
Yeah. Absolutely. Absolutely. One, I think five to ten years definitely gives us time to put some drugs onto the market, which is definitely our goal. I mean, most of us are here because we want to see more drugs developed, and we've got some sort of personal connection to see how that would come through. So, that is what drives us every day. I think five to ten years from now, we will have either changed the system and be a leader in it ourselves, or we will have changed the system and someone else will be leading it. And so, either way, I will feel better about that. I think it's funny because five years ago, when we were doing this, nobody talked about the impact that we could have, right? None of the large pharma talked about their own AI development programs.
Just so many of the things that are now commonplace, more than three quarters of biotech now incorporate some form of AI into how they're developing it. We're just getting better results. So, I look forward to seeing the entire industry shift. I hope that we are the leader that drives it.
Great. Well, thank you very much for taking the time. It's been an absolute pleasure. So, thank you.
Thank you.