Recursion Pharmaceuticals, Inc. (NASDAQ:RXRX) Q4 2023 Earnings Call Transcript

Recursion Pharmaceuticals, Inc. (NASDAQ:RXRX) Q4 2023 Earnings Call Transcript February 27, 2024

Recursion Pharmaceuticals, Inc. isn’t one of the 30 most popular stocks among hedge funds at the end of the third quarter (see the details here).

Chris Gibson: Hi, everybody. I’m Chris Gibson, Co-Founder and CEO of Recursion, and I am really excited to welcome you to our first ever Learnings Call here at Recursion. So what is a learnings call and why are we starting this practice now? A traditional learnings call has a lot of value, but over the years I think these have become extraordinarily scripted, frankly quite boring in many cases and hard to access for all of the stakeholders that we want to be able to speak to. Learnings is our interpretation of a traditional earnings call, which we feel is more authentic, so I will not be scripted today, I’ll just be working off of the slides in front of me, adaptive and we hope easy to access. And please, if you have suggestions on how we can make this better going forward, please send them our way.

What I would also say is that, we’ve chosen to initiate our first learnings call at this moment, at the start of 2024, because as we look ahead at the future of Recursion, the milestones and catalysts coming before us are going to be coming fast and furious, and we want to make sure that we have a robust mechanism to reach out to all of our stakeholders on a quarterly cadence, and to be able to share all the incredible work that we’re doing here at Recursion with you. So to frame where we are today, where we’ve been and where we’re going, I want to start by going back really a decade, going back to the origins of TechBio, one decade ago. And it was a really interesting time in the early 2010s. You saw technology companies coming into a wide variety of industries and leveraging a pretty straightforward playbook to bring fundamental new advances from how we get around cities, to how we think about our preferences for digital media, to how we even think about what products we want to order.

And what these companies did was quite straightforward. They used technology to capture high dimensional data to create a digital record of reality. And it’s important to note that the data that they collected was rich, very, very rich and high dimensional. They aggregated and digitized that data, and then leveraged algorithms to make predictions across all of these massive data sets. And most important of all, they went back into the real world to test those predictions. So whether that’s telling you to turn left instead of right, whether it’s telling you to buy product A instead of product B or to watch TV show X or Y, these algorithms could be tested in their ability to predict the right outcome in a real setting. But in biology, this has been extraordinarily challenging.

There are so many roadblocks to aggregating and generating the right data to be able to map and navigate this complex system of biology and chemistry. There are three primary drivers of that. First, this world is very analog standard. It was more so in the 2010s, but it still is in some ways today. There are still CROs who send you scanned PDFs or printouts with handwritten notes. And in the biopharma industry, there’s a tremendous amount of data, hundreds of petabytes of data, but that data was collected in a way that wasn’t built for the purpose of machine learning. And so it’s often siloed on legacy servers, it’s often built without the right kind of high dimensional nature or the right kind of metadata to make it easier to extract the connections across and between all of those different data.

And then of course, there’s the public datasets that we and others use. But as you all know, there’s a reproducibility crisis, and there are real challenges, because just like in the pharma data, there’s not enough metadata and not enough relatability of this data across all these different publications and data sources. And so it’s very, very challenging in the biopharma industry to aggregate and generate the right data. But what we and other companies who are today leading TechBio saw in the early 2010s was an opportunity. We saw exponential improvements across five main areas. The first was the cost of storage. So in the early 2010s, we were at the end of a 40-year cycle of precipitous decreases in the cost of storage. And this is important because a company like us at Recursion today with over 50 petabytes of proprietary data has to be able to pay to store all of that data.

See also Top 20 Most Valuable Tobacco Companies in the World and 15 Countries with Most Breast Implants in the World.

Q&A Session

We were seeing a radical increase in the availability of compute. We’ll talk more about our supercomputer a little bit later. We were seeing an increase in accessibility and flexibility of automation tools that allowed us to pioneer and industrialize a new kind of omics using robotics. We were seeing a renaissance in new biological tools like CRISPR. And then, of course, the field of AI was making extraordinary strides as we took 20 years of learnings and really invested in billions of dollars across the tech industry to move from expert systems into this neural net modern AI age. And now fast forward to today, where Recursion is right now, leading TechBio. We are taking that same formula that was so obvious across the technology companies of the early 2000s and 2010s, and deploying it now across the biopharma industry, where, as I said before, the data is so hard to generate and so hard to aggregate.

But we are doing it here at Recursion. We’ve built a massive automated platform where we can profile biology across human cells, rodent cells, in vivo systems and even patient data. We can extract that data in high dimensional space, aggregate it and then train algorithms on our supercomputer and cloud computing resources to make predictions. And this is the most important part. More than any other company in this space, I believe we are set up to take the predictions from our algorithm and test them back in the lab and creating that virtuous cycle of learning and iteration is the Recursion OS. It’s what we’ve been building for the last decade and it’s what we see positions us, the data, the technology together and this virtuous cycle to really define and lead the TechBio space in the decade going forward.

But we’re not just building at one point in the drug discovery and development process. It takes hundreds of steps to discover and develop a drug and Recursion today is building these virtuous cycles of wet lab and dry lab of learning and iteration at points from how we connect patient data into our targets to how we optimize chemical compounds, how we translate these programs and now early work in how we even identify the right patient cohorts to drive our programs into the clinic. I think more than any other company in this space, really building the full vertical TechBio solution. And that means that we are leading TechBio in 2024 across three primary areas. Our internal pipeline, our partnerships and our platform. Recursion is leading.

Our first-generation programs, five Phase 2s, either enrolling or soon to enroll patients that are really focused in capital efficient niche areas of biology. And we’re excited to have second-generation programs that are leveraging some of the tools that we have built or added to our platform in just the last few months moving to the clinic as well. If we build this platform right, every generation of programs will be better than the last. But it’s not just our internal pipeline. We’re also learning from and working with partners across both bio and tech. On the biology side, we’re partnered with Roche-Genentech in neuroscience and one oncology indication and then also partnered with our colleagues at Bayer in precision oncology. But unlike many other companies in this space, we not only have the therapeutic partnerships, we also have partnerships across data with companies like Tempus, across compute with companies like NVIDIA and across chemistry with companies like Enamine.

And it is this cross-credentialization of technology partnerships and biology partnerships that we believe sets us apart. And all of these partnerships and pipeline are based off of the Recursion platform. Today, over 50 petabytes of proprietary biological and chemical data spanning human cells to rodent cells to model organisms to human patients. And in order to make use of all of that data substrate, at Recursion today, we now own and operate the fastest supercomputer in the biopharma space. And in order to take the predictions from the algorithms that we generate on this computer and test them in the lab, we have industrialized and automated multiple levels of omics data generation at Recursion. On our Phenomics platform, for example, we’re able to do more than 2 million experiments in any given week.

And so before I talk about what we’re looking out to in terms of our near-term catalysts and milestones for Recursion, I want to take a moment to just look back at 2023. And I want to do this because I think it was one of our very best years. Amidst a challenging capital market environment, this team delivered on our pipeline, our partnerships and our platform. And so we’re going to go through just a few of the highlights. First, I’m going to start back in May, where we announced simultaneously on the same day the dual acquisitions of Cyclica, a digital chemistry company that’s based in Toronto, and Valence, a cutting-edge AI laboratory for drug discovery that’s based in Montreal. And we were able to fully integrate the Cyclica team in just 90 days.

And in a few minutes, I’ll share with you some of the output from that acquisition that led to us advancing and improving our programs within just a few months of signing that deal. On the Valence side, I’ll show you LOWE later, which is our Large Language Model workflow orchestration engine and this has really been driven by the Valence team. And I will set the stage for how we see a new direction for how biopharma is going to access all of these incredible new TechBio tools. In June, we announced that our first clinical trial, SYCAMORE, this is a trial for the first therapeutic candidate to be advanced by any industry sponsor into Phase 2 for cerebral cavernous malformation and I will remind you that is a massive area of unmet need. This is a disease that affects roughly 6 times the number of patients as cystic fibrosis and yet we are the first with an opportunity to be first in disease.

This program was fully enrolled in June across 62 patients in three arms. And one thing that gives us a lot of confidence about the tolerability of this molecule is that today, as patients finish their 12 months on therapy, the vast majority continue to opt into our long-term extension study. And so we’ll be reading out the topline phase two data in Q3 of this year. This will be our first real POC readout and we’re really excited about the opportunity, not only to potentially drive forward an exciting medicine for an area of significant unmet need, but also regardless of the outcome of that study, to learn and put that data back into our platform so that the next generation of molecules can be even better. Then in July, a month later, we announced our collaboration with NVIDIA.

This included a $50 million equity investment. And with our partners at NVIDIA, we’re working on advanced computation, so foundation model development; we’ve got priority access to compute hardware, which I’ll talk about later; and the DGXCloud Resources and we talked with them about the potential for us to put some of our tools into their BioNeMo marketplace. And in fact, just last month in January at the JPMorgan Healthcare Conference, we released the first third-party tool to exist on NVIDIA’s BioNeMo platform. That was our Phenom-Beta foundation model in January of 2024. So very excited about this ongoing collaboration. One month later in August, we were able to deliver a demonstration of how we leveraged the May acquisition of Cyclica and our brand new partnership with NVIDIA to drive a real value into our platform.

We were able to predict the protein ligand interactions for more than 36 billion compounds from the N Enamine REAL Space across about 80,000 predicted binding pockets spanning the human proteome. And what this did was generate a large in silico data layer for us, a synthetic data layer. So when we find a new target or an initial hit, we can immediately prioritize that target based on a potential mechanism of action and we have already advanced multiple programs, terminated multiple programs or changed the course of multiple programs using this exciting new technology. So we really see it as fantastic to have the complementarity of this functional machine learning algorithm alongside our or this physical machine learning algorithm alongside of our functional biology based platform here at Recursion.

Then in September, we announced the Phase 1 study results for REC-3964 in C. diff colitis. The molecule was safe and well tolerated at multiple doses up to 900 milligrams. There were no SAEs and no discontinuations that were related to treatment. And along with the favorable PK profile, this gave us the confidence to advance this new chemical entity towards a Phase 2 trial, which we will initiate later in 2024. Then back to our platform in September, we announced our first foundation model we call Phenom-1. It’s the world’s largest Phenomic foundation model that we’re aware of. And I want to take a moment just to talk a little bit about this, because I think it’s really exciting, especially given all of the talk around Large Language Models in the background.

In a Large Language Model, one trains a neural network to predict the next word in a sentence or in a paragraph. And we’ve done something similar here, but instead of using written language, we’re using the language of images of human cells. And what you can see on the left is an image where we’ve masked 75% of the cellular image and we’ve trained a neural network to predict what the rest of that image would have looked like. That’s the middle row here and you can — or the middle column. And what you can see is that our neural nets got really good at doing this. You can almost not even tell the difference between the Phenom-1 reconstructions and the original image. But we’re not in the business of reconstructing masked images at Recursion.

That’s just a training task. And like in a Large Language Model, where the ability to predict the next word in a sentence led to these emerging features that almost gave us a sense of rational thought in ChatGPT and other sorts of settings, we’re seeing emergent features from these foundation models. So against a wide variety of benchmark tasks in drug discovery, these sorts of models are giving us state-of-the-art performance to rediscover known biology, to make predictions about admin talks and beyond. One of the things that was most interesting about this work, though, was that we were able to demonstrate that the scaling hypothesis holds in the world of biology. We were able to demonstrate that is that the bitter lesson holds true and that one must have more data and more compute, all else being equal, in order to build a better model.

And so based on that, just two months later, we announced with our partners at NVIDIA that we were expanding our supercomputer, which was already the fastest supercomputer wholly-owned and operated by any biopharma company, with another 504 NVIDIA H100s. And this is a picture of the team just a week or two ago where these H100s have arrived on site and we believe when this system is up and running, it will not only be the fastest supercomputer in the biopharma space, but it has the potential to be one of the fastest supercomputers privately run in any industry. So we’re really, really excited about the potential to get this thing up to speed and humming. But going back to our partnerships, in October, we also announced that Roche had exercised the first program under our collaboration, this program in the context of oncology.

And this was fantastic, less than two years after signing that collaboration to already have a program advancing forward with our partners and we hope and expect that this is the first of many options to come across this partnership and others. In November, we then announced another partnership. This time, instead of just generating data at Recursion, partnering with Tempus to aggregate what we believe is extraordinarily high quality patient data into our platform, access to the DNA and RNA sequencing data sets and clinical records for over 100,000 patients that we can now train causal AI models on using the Recursion OS. That gives us now access to over 50 petabytes of proprietary biological and chemical data that we’ve either generated in-house or partnered with companies like Tempus to bring in place.

And I’ll talk more in a minute about how we’re already leveraging this partnership to drive value in our platform. Also in November, we announced an update to our partnership with Bayer, focusing on precision oncology. And I think it’s important to note that with this update, we were able to more than double our per program milestones, which I think is a strong signal about Bayer’s excitement around what we’re building. And I know the teams are already hard at work together at Bayer and at Recursion to drive forward some of these initial new oncology programs together. Coming out of that same partnership, before it moved to precision oncology, it was focused in fibrosis and there was a program that was part of that that we thought was just too good to let go to waste.

And so we were able to negotiate with our colleagues at Bayer to in-license this program, which we call Target Epsilon, which we believe is a novel target in the context of fibrosis and we are driving this program forward very quickly. In fact, we’re announcing with today’s earnings that this is now in IND-enabling studies at Recursion. So we’ve already advanced it inside of our own internal pipeline. And finally, in December, we also crossed the threshold of having generated over 1 trillion neuronal iPSC cells since 2022 and based on the publicly available data, we believe that this makes us the world’s largest producer of high-quality neuronal iPSC cells. And this is but one example of the way our team is working with complex biology, co-culture systems, a wide variety of biology to drive our platform forward into new, exciting areas like neuroscience.

All of this underlying our pipeline, which, as I shared earlier, we believe is the most robust, deepest and broadest in the TechBio space. And we are now looking forward at 2024 with this learnings call setting the table for a number of important catalysts that are coming up. First, our Phase 2 topline readout for CCM in Q3, then a preliminary safety and efficacy readout for NF2 in Q4, and then in the first half of 2025, a preliminary safety and efficacy readout for FAP, the initiation of our Phase 2 program for C. diff colitis later in 2024, and then another Phase 2 safety and preliminary efficacy readout in the first half of 2025. So Recursion really beginning with this third quarter in 2024, setting the table for what we hope can be roughly quarterly readouts that we hope will help propel the company and the platform forward.

Beyond these early first-generation programs, we’ve got our Epsilon project and our RBM39 project, which are the first of our second-generation of programs, making use of some of our newest tools and we’ve got more than a dozen discovery and research programs in oncology or with our partners coming behind those. Now, before I talk about where we are today and what we see as catalysts in the near-term beyond just our pipeline, I want to orient you to the broader trajectory of the space of TechBio, at least as we see it. And to do that, I have to go back a ways again, to the early days, back to the 2010s, when companies like Recursion were founded. And all of these companies really made their start with a point solution and we’re no different.

We were scaling, industrializing, and pioneering a new kind of omics based on images of human cells to try and understand and explore biology. And since that time, we’ve actually seen that our work in this space has just continued to grow in complexity. Today, we can leverage our automated platform on Phenomics to generate more than 2.2 million experiments worth of data every week. We leverage extraordinary foundation models like Phenom-1 that I talked about earlier to make predictions about the relationships across more than 5 trillion biological and chemical contexts. This is an extraordinary, extraordinary feat, and it’s based on broad biology, over 50 human cell types that we’ve explored, roughly 2 million chemical compounds, whole genome CRISPR knockouts.

This is really, really exciting work that we continue to push the limits of. But this is but one step in the Recursion OS today. While we started with Phenomics, it is now one of many steps spanning patient connectivity all the way to the clinic. And while I wish we had time to go through each one of these, I’m just going to focus on a few of these areas that I think are important to illustrate some of our focus on building these virtuous cycles. And the first of those is DMPK. Our DMPK platform is now up and running at Recursion. This is a highly automated platform that’s allowing us to execute three critical assays across both human and rat contexts. We can do nearly a thousand compounds a week on this automated platform and this is great because we can profile the molecules that are moving through our internal pipeline or our partnership pipeline.

But what’s more, we’re using the majority of this platform’s bandwidth to actually profile many diverse compounds to build the data substrate on which we can train additional state-of-the-art predictive ADME and Tox models. And it’s this virtuous cycle of learning and iteration, of data generation and algorithm improvement that we think will differentiate us not only in target discovery with Phenomics, hit discovery with Phenomics, but even in how we advance our molecules towards the clinic. And it doesn’t just stop in human or rodent cells. We’re building these same kind of tools in model organisms. In our vivarium, we have over a thousand cages with cameras and other sensors that allow us to extract much richer, high-dimensional data from each one of these animals.

And this means we can use fewer animals as we drive our programs forward and it means we can make decisions in real time. We can deprioritize and prioritize molecules based on digital tolerability studies in real time and this has already made a difference in both accelerating and leading to the faster termination of programs at Recursion. But it’s beyond model organisms. It also goes to the ultimate model organism and that is humans. With our Tempus data, we’re able to now aggregate patient data across oncology together with all of the wet lab data we’ve generated at Recursion. And in just about eight weeks since we’ve had access to this data, this has already led to our team combining our wet lab data and the patient data. So forward and reverse genetics coming together and allowing us in the context of non-small cell lung cancer to already identify multiple potential drivers of disease that we are predicting are causal, which in many cases have not yet been robustly explored in this space.

So Recursion now has a program that has advanced just in the first eight weeks based on this kind of data and we’re really just getting started. But what’s happening is that as we continue to build this full stack of technology tools and as each of these tools runs through its virtuous cycle of learning and iteration and is improved rapidly, it’s becoming increasingly complicated for anyone to keep up with the latest on each tool, the right way to use each of these tools and we actually think this is going to be a problem across the industry, as we and many others are building lots of models and lots of different tools. And so we wanted to address that together with our colleagues at Valence Labs. And we were able at the JPMorgan Healthcare Conference, both in the conference, we think for the first time doing a live software demo and also at the event we co-hosted with NVIDIA to show off our LOWE system.

This is a Large Language Model-Orchestrated Workflow Engine. And what this is allowing you to do, what our scientists and our partner scientists may be able to do with this technology, this tool, is to use natural language, to not have to be an expert programmer, to be able to access all of the tools, to be able to design experiments the right way, to order experiments and execute them on our platform, to analyze data and visualize data using the latest tools at Recursion. And really, this kind of technology is putting the power of the Recursion OS at the fingertips of all of our scientists and partners. And we see this trajectory as very similar to the early days, the late ‘70s and early ‘80s, in the personal computer space. You had products like the AppleOne on the left, where you really had to be an expert user.

You had to be comfortable with this microprocessor board. You had to be comfortable working at the command line in order to make use of this burgeoning new technology. And with subsequent Apple models, including Lisa on the right, we moved to a graphical user interface. And this created really a renaissance in the ability of more people to be able to harness the power of compute. And what we’re building with LOWE, with the Recursion OS, we believe is akin to this, but it’s really a discovery user interface. And we believe it’s going to allow each scientist at Recursion and beyond to make more progress faster. It’s going to mean that our teams are doing less of the toil and more of the thinking around our projects and it also means that these tools are going to be accessible, not just to scientists in biology and chemistry, but to software engineers and data scientists, to BD and to finance.

And we think ultimately that’s going to be fantastic for the field and we believe Recursion is really leading out on this new trajectory for our industry. So before I move to questions, I want to just end with our near-term milestones, the things that we believe we’re going to hit over the next 12 months to 18 months or sooner. And I’ll start with additional INDs. We’ve got both our RBM39 program and our Target Epsilon program that we in-licensed from Bayer, moving towards the clinic. We’ve got more Phase 2 trial starts, AXIN1 or APC and C. diff that we believe will be starting this year. We have multiple Phase 2 readouts that I alluded to earlier. And all of this on top of a healthy balance sheet with nearly $400 million in cash at year end 2023.

And what’s more, we see the potential for significant runway extending options for our map building initiatives with partners and for additional partnership programs being optioned. And beyond that, we see the strong potential for additional partnerships in large intractable areas of biology, like cardiovascular metabolism and immunology, where we expect robust upfront payments that will further extend our runway. And what’s more, we have an ATM open, which we’re using in a very, very surgical way with the right investors at the right time in order to make sure that the company maintains a robust runway moving forward across all of these exciting catalysts. And finally, we’ve got the potential both on the BioNeMo platform and through our LOWE tool to make some of our data and some of our tools available to biopharma and commercial users.

And there’s the potential for some of that work to generate additional revenue as well. So I hope you’re as excited about the future of TechBio as I am. I hope this has been helpful for you to see the trajectory of the company through 2023 into 2024 and how we see the future of our industry. And with that, I’m going to stop here and head over to answer some questions. And these are being updated by our team live. If you haven’t had a chance to ask a question yet, please log into the Slido tool and do so now.