Welcome to Raise the Line with Osmosis from Elsevier, seeking solutions with leading experts on how to increase healthcare capacity so people can get the care they need during the pandemic and beyond.
Hi, I'm Shiv Gaglani. We've been talking a lot lately with family members of people with rare diseases about the need for more rapid development of drug therapies. Well, today we're going to take a look at a company whose mission is to create a more efficient path to drug discovery, and it's using one of the most powerful supercomputers in the world to do it. I'm happy to welcome Chris Gibson, the co-founder and CEO of Recursion, which in its nine-year history has created more than a dozen preclinical and discovery programs in diverse therapeutic areas. Chris developed the technology and approach that seeded Recursion as part of his MD-PhD work while at the University of Utah. After completing his PhD, he left medical school to launch the company, something we have in common. So, Chris, thanks for taking the time to be with us today.
Thank you so much for having me. I'm really excited to be here.
So, I've been a fan of Recursion ever since I heard about it from our shared investors, Felicis Ventures, which before the show started I mentioned led our Series A at Osmosis. You know, you obviously have a very impressive background, and the more I've peeled back the layers of Recursion, the more interested I've become. So, for our audience, many of whom want to follow in your footsteps and maybe create companies or join exciting companies, can you tell them a bit about what got you interested in bioengineering and then drug discovery and development in the first place?
Yeah, I mean, I would have to go so far back, back to the sixth-grade science class. But you know, I always loved science from the very beginning, and I had a teacher in sixth grade, and our final project at the end of that year was a thing called sludge, where we got a jar full of a bunch of different, you know, materials, and our job was to use everything we learned over the year to separate those materials, figure out their density and all this sort of stuff, and identify them. And I picked the hardest sludge that one could pick. I think it had like 12 things in it, including gases that I had to capture. And it was a miserable failure. I think I got like three or four out of the 12 things right.
But ultimately, it was trying to understand something super complex like that using kind of an engineering and scientific approach that just spoke to me. And that teacher actually took me to lunch because I was pretty upset with how poorly I had done. And she just somehow turned some switch inside of me that going after these big hard things is worth doing. And it was, I feel like at that moment, that was the moment I look back on as the start of Recursion in many ways. And it started this kind of insatiable hunger for science that I've been following ever since. And I tended to operate mostly at the intersection. I loved that about bioengineering. You weren't going all the way to the full depth of any engineering discipline or in biology, but trying to understand how lots of different fields kind of interacted. And that's just always been a place that I think is a fascinating place to be.
That's fascinating. I mean, I love that story for many reasons. And one of which is that, you know, we're an education company at Osmosis, and many of us got into science because of those influential teachers. And you're telling me this was sixth grade, this teacher had you guys do this?
Sixth grade, Carol Pongonis. I still remember it. And I think all of us have teachers like that in our lives.
That's awesome. Yeah, we just wrapped up this thing called the Raise the Line Faculty Awards to honor teachers like that. For me, it was Sherry Jenkins, my anatomy teacher back in ninth grade, who got me really interested in this stuff. Does Carol know what you're doing now? I'm curious if she's followed your career.
She does. We spoke a few years ago, and I try to make a note of following up with her every couple of years.
That's awesome. Well, certainly I'm going to go on more into questions, but one last question on this sludge concept because I find it fascinating. Did you see the recent research report by Meta where they took samples of dirt and they found, using AI, they were able to basically find the structures of tons of proteins that we've never even seen before?
Yeah, I think it was like 600 million or something like that. It's incredible. And it's interesting too to see, you know, TikTok had a job post up recently for a digital chemist. DeepMind with Isomorphic going into the space, Meta getting into the space. There's some really interesting technical challenges that are in the field of biology that seem to be a really good fit for some of the most advanced computational things that humans have built over the last decade or so. So, it's going to be fascinating to watch the intersection of those fields in the coming years.
Absolutely. Yeah, AlphaFold, all these different amazing things that are kind of the basis for a lot of the innovations that you all are leading. So, let's actually go from sixth grade to Recursion. Tell us a bit about your MD-PhD and your decision what to do the PhD in, why leave medical school to do this, and then we'll get into Recursion.
Absolutely. So, I joined the lab of a guy named Dean Li when I started my MD-PhD at the University of Utah. And Dean is a physician-scientist. He's actually the president of Merck Research Labs now, so a very translational guy who's gone on to the pharma industry. And his lab brought together folks from a lot of different backgrounds to study mostly the idea of vascular stability, how our blood vessels are leaky or not leaky, and the role that that plays in lots of different diseases. And one of the diseases we were studying was one called cerebral cavernous malformation, or CCM for short. And it's a rare genetic disease, actually a pretty common rare genetic disease, about six times more people than cystic fibrosis, but one that hasn't gotten the same kind of research coverage as CF.
And we were trying to understand that disease as a genetic model of vascular stability, vascular instability, because people basically get, you can think of it almost like mini aneurysms in the capillaries of their brain, and these aneurysms get sort of leaky. These lesions get leaky. So we were understanding the genetic underpinnings of that disease. This started well before I joined Dean's lab that built animal models, helped identify some of the genes that caused the disease, and it helped dissect the molecular and cellular biology that they thought underpinned the disease, and I got to help with the end of that. And we were determined to figure this thing out, and we figured out and published that it was probably driven by activation of RhoA. That was what some of the pathophysiology was driven by. And so we decided to follow that up in the mouse model and inhibit RhoA with simvastatin, because we're an academic lab, we weren't looking to make drugs at the time, through HMG-CoA reductase. And we did, and we unveiled the data, and we made the mice worse after five months of treatment.
And it was this humbling moment of biology for myself and I think many people in the lab, where the overwhelming complexity of this space we operate in, the space of biology, was so complex that the best kind of molecular and cellular biology tools in a lab that was regularly publishing in Nature and other high-end journals was pretty wrong. At least at the time, the data suggested we were wrong.
And that was the moment where I think maybe Recursion really started a second time. And over the coming weeks and months, going to various lectures, including one by Stephen McKnight from UT Southwestern, there were a bunch of little pieces of data that got us excited in his lab about doing a phenotypic screen so that we could cast aside some of our bias about what was driving this disease. Just model the genetic loss of function because we knew that incontrovertibly to be true, and then ask the cells what was actually driving the disease and what could make the disease better. And so we built a phenotypic screen where we used microscopy images of human cells where we'd knock down one of the CCM genes called CCM2 with siRNA, and the cells looked radically different under a microscope. And rather than use a hypothesis-based approach, we just added thousands of known bioactive molecules.
And used actually a computer vision approach pioneered by a woman named Ann Carpenter at the Broad Institute called CellProfiler to use the computer vision to train a very basic machine learning classifier to recognize the disease cells and the healthy cells. And we picked out about 39 drugs or molecules that at the time made the computer vision algorithm think that these cells looked healthy if they had started as diseased and we added some drug. And we went through a variety of other assays and eventually took two of those drugs into the animal model again. This is probably a year or two after that failure. One of them was vitamin D, and the other one was what is now called Rec-994 that we have in Phase 2 trials for CCM. Both of those molecules made the animals better. And so this traditional kind of hypothesis-based reductionist approach led to a failure. This hypothesis-agnostic, unbiased approach led to an animal-based success, and that's where we asked, could we go do this 50 times, 100 times, 1,000 times and replicate this approach to explore biology more broadly.
That is fascinating. Thanks for breaking that down for us. I mean, I think a lot of our learners probably have been in wet labs, you know, whether they're undergrad or high school doing research, maybe now in med school or nursing school. But to know the hop, skip, and a jump, it's quite a big leap actually from maybe their experience in wet lab doing this hypothesis-driven testing to the scale and the way you've used everything from in silico biology to high-throughput drug screening, etc., to do this. So, can you give us a bit of a sense of Recursion as it is today? You know, what are some of the things you are working on that get you excited, and what is your scale? If you can do anything like I mentioned in the intro, one of the best supercomputers in the world, talk to us a bit about that, too.
Happy to share. Happy to share. I mean, for clarity, when we were working on CCM, I was using a multi-channel pipette to do all that work. So, it was relatively low throughput. I joke that today we do the equivalent of my entire PhD's worth of experiments about every 15 minutes at Recursion. But that's based on this giant automated laboratory that we built, which is just, you know, 100 feet away from me now, full of robots where we can do today up to about 2.2 million experiments a week in our image-based approach. So, this is microscopy images. We generate more than 10 million of those images a week across more than 2.2 million wells is our kind of capacity. And of course, it takes several people to operate that lab. We actually run two shifts six days a week. But still, for a relatively small number, about a dozen people to be running that lab and generating that scale of data is pretty incredible.
But we've moved beyond just that image-based approach. As you mentioned, we now have built our own small molecule library at Recursion where we have nearly 2 million molecules physically in-house. That's kind of the same scale as many of the large pharma companies, and we can screen these molecules in various disease states. We have, as you mentioned before, we operate one of the 500 fastest supercomputers in the world. It's called BioHive-1. It's a 40 DGX-100 supercomputer system. We use that to train neural nets on all of this data that we generate. And there's some specific technical reasons why we do that on-prem instead of in the cloud for training of our neural nets related to kind of the size of the images that we use.
We also have started scaling other areas of biology. So, we do transcriptomics experiments at Recursion to the tune of about 13,000 of those samples a week. We have a facility where we do animal work, and that rodent work is done with cameras in the cages that allow us to use machine learning to visualize how the animal is interacting with its environment, interacting with its neighbors, interacting with molecules that we might use to try and ameliorate a disease model. And so, really, Recursion, everything is about scale and then using technology to measure really, really broadly these high-dimensional signals and to get away from the traditional Western blot, right? Where you're kind of measuring one thing and it either goes up, it goes down, or it stays the same. You can trick yourself a lot of ways with those low-dimensional readouts. When you're measuring thousands of features in biology and chemistry, I think it's a lot harder, as long as you do the statistics right, to convince yourself and trick yourself that you're accidentally moving so many different features in the right direction.
So, we really do operate at massive scale, and today we've generated almost 20 petabytes of proprietary biological data. And to give you a sense of scale, if you took every feature-length film in human history in every language in 1080p, we would have about 10 times that amount of data here at Recursion. So, it's a lot of data.
That's incredible. But I love how you communicate, too, because it's very analogous to what Osmosis is, like the largest health education channel on YouTube. A lot of analogies and kind of, you know, the 15-minute PhD is very interesting. It's to me, I'm sure some of our listeners who are doing their MD-PhDs wish that they could do their PhD in 15 minutes.
So, you know, in this era of high-throughput, and, you know, I know Illumina just announced that they have a $200 whole genome sequencing available now, trying to get to that $100 mark or some $100 mark. The challenge becomes finding the signal from all the noise and taking these petabytes of data and like finding those things that could work. How do you, you know, talk us through how do you go from high-throughput to then actual clinical trials and what's Recursion's involvement once you get to, you know, Phase 1, Phase 2, Phase 3? Do you hand it off like these promising targets and promising molecules to like a larger pharma company to take through completion? How do you see it?
