I think we're right about at 2:10, so I think we'll get started. I'm sure you've all heard this a couple times today, if not this week, but thank you again for coming to TD Cowen's 44th Annual Healthcare Conference. I am joined today by the CFO of Exscientia, Ben Taylor. Thank you for coming. Thank you. Really a pleasure to be here. Awesome. So look, I think AI is in every headline, you look left and right, in every industry, every sector, impacting our lives probably more than we all want to admit or realize. But maybe give us a quick sense, first at a high level, and we'll kind of dive into, you know, the applicability in each of the different verticals here. But really, what is it about your platform? Maybe try to explain it to us in a way that maybe some of us who aren't super steeped in all the AI- No, of course. knowledge can really just kind of understand where and how you match up to some of these other platforms. Yeah. So I think there's a couple of guiding things that we do that are just a little bit different. So, one aspect, we're not a big data company. We do generate a lot of our own proprietary data, but it's all for very specific purposes. So a lot of times you'll hear about AI companies, and they are basically looking for correlations in data to try and find a question, I guess, is a way that you could think about it. For us, we generally know what the question we want to answer is, but we're in a sparse data problem. There isn't an existence of data that we can search through to find the answer. So we have to figure out what is the most efficient system to actually work through and get to the answer for that. And so if you don't think about AI as big data, you actually just think of it as a logic stream. That makes a lot more sense. We can actually do multiparameter optimization of a whole bunch of different logic streams to get to a better end result. And that's different than most people are used to with AI, but we're doing a very different business. The other aspect I'd say is we really try and integrate the virtual with the experimental. So half of our company are coders and data scientists, half of them are actually producing experimental data. And so that's a big difference. We figured out about six years ago that if you were virtual only, you never could actually evolve past what the CROs could do. You are always gonna be limited to their timescales, their cost, and their data quality, which often isn't as high as we would want it. And so we started to build our own abilities to do assay development and all of the different aspects around how you define a problem. I think that the easiest way to think about it is you have to know what good looks like, and you have to be able to produce that data reliably so that your AI systems can find an answer for that problem. And so we have very specialized data production that changes based on every problem that we're trying to solve. So that's another bit of a different way to think about it. The final thing I'll say is, we have a lot of different technologies, but they all link together in one way. So there's not a single algorithm that defines us. There's not a single part of the process. What we actually do is we say, "For a particular drug, why is it most likely to fail when you're going through the clinic? Or what is going to limit its commercial opportunity? And can we create a predictive model that actually tells us how to design a better drug to prevent that?" And so everything that we do, and we have hundreds of algorithms, we run every virtual molecule on thousands of different models. We do all sorts of different things, but it's all with this intention of: we think it could fail later on for this reason, and we're gonna create a model to figure it out before it's even a drug, before it even hits a patient. Okay, so that's great. That's a great segue into, I think, diving a little bit deeper. I guess what's often interesting, the longer you stick around some of these biotech companies, who have their own type of AI platform or who integrate them later on, it's always interesting to see how these things evolve. Mm-hmm. Inevitably, they will, as any company will. But maybe give us a sense of what are in your experience with the platform, what are kind of the more important pivot points that you guys have focused and leaned into? And on the flip side then, what do you foresee as kind of what are the next steps and goals that you want to achieve with it over the next, let's just say, year or two? Yeah, so one of the big shifts for us was realizing generalizable models don't work well- Mm-hmm. In specific drug development. And so we were using generative AI back 12 years ago, and we used a lot of different model systems to try and say: "Okay, for a random drug target X, let's just quickly put something out because we've got generalized model systems that can do that." And what we found is, we got generally okay results. And so what we needed to do was really go in specifically to each situation and really create a design environment and data production for that specific environment. And so that's been something that we've really leaned in a lot to. We do use generalizable models, we do use LLMs, we do use different things like that. But I would say the design environment is uniquely set up for every single project we do, and I think that's why we've just had a lot more success than other people have. I mean, we've had eight development candidates. We've put five or six into the clinic or getting ready to go into the clinic now. And those are really specific, measurable outcomes where we've been able to design solutions to problems that traditional methods weren't able to do. Okay. So when you kind of look at the functionality of the platform as it exists today, as it's come to, and really where you wanna take it, are there certain therapeutic disciplines that you're driven by? Like, you particularly want to design it in a way that's maximized to, you know, have a really great oncology drug or something like that. Is that part of the strategy, or is it a little bit agnostic at this point? Well, if you think about the two different sides to the problem, the chemistry and the biology. So, the chemistry is something where we sort of don't care what the therapeutic area is. We still have to look at the target very specifically. We still have a lot of specific design problems, but it's when you step back and you look at the biology, then all of a sudden you need a lot of therapeutic area expertise. You need to know, you know, how the clinical trial is going to be run. You need to know what standard of care evolution is. Mm-hmm. You need to know, you know, all of the different mechanistic aspects. And so what we've done is we have two different business models. They all use the same chemistry design platform, but for certain ones in oncology, we actually do our own translational science, also using AI and different methods. We've built our own clinical development team and have our own KOL networks and all of those aspects for oncology specifically. Mm-hmm. Outside of oncology, that's where we're reliant on a partner to come in, and so you saw us license one of our discovery compounds into the Sanofi collaboration, just before the end of the year, and that was because it's an I& I target. Mm-hmm. It's a really beautiful molecule. We look forward to talking about it, sometime in the near future. But, we are not set up to do large-scale I&I indications Mm ... from a clinical development perspective. And so Sanofi is, they're a world leader in it. We'd much rather partner with them. We can do the parts of it where we add value on designing out problems that other people weren't able to do, and they can do what they do really well, which is take it forward on the development. Okay. All right, great. I mean, I definitely want to kind of nail a little bit down into your existing partnerships and then really your strategy for future ones. But I guess maybe before I do that, I know there's kind of this precision medicine aspect to the- Mm ... the whole platform as well, and then really how it integrates with your, you know, your automation, new automation facility. Maybe just help us understand what that realistically means for you. Yeah. How are you integrating actually some of these precision medicines when you're developing a target, when you're deciding how to develop a target, and then really how that relates to, you know, your next steps with the platform? Yep. And we use precision medicine a little bit differently than I think is commonly, Mm ... thought of, and that's probably our own fault for not coming up with more inventive wording. We don't mean it to be limiting down to really narrow patient populations. Mm. What we actually mean by it is, how can you best define your patient population so that you can design something very specifically for them? And so that can be a very large patient population. Like, if you look at CDK7, that is a fundamental cellular mechanism. However, we can still differentiate some patient populations respond better to it than other patient populations. And so even though that some patient group is actually a majority of, advanced cancer patients, it's still precision medicine in our mind because we're actually defining these patients. Before, a lot of times, what we've seen in clinical development is people are using clinical trials as exploration of a compound and exploration of a patient population. They should be confirmatory, right? You should have designed a compound that is saying, "I'm gonna go after this specific patient population, and now my clinical trial is confirming that I was correct on that." And so that's what we mean by precision medicine. Now, what we've done is, in oncology specifically, we have a patient tissue platform. So we take live patient samples, and we're able to put them into a well-based system. We keep their three-dimensional nature. We don't amplify. We don't do anything like that. We keep an immune compartment, which is really important because what we want to see is how all of those different cells interact when they're exposed to a certain drug. And we can use AI then to actually measure every single individual cell's reaction. So we can see how the B cells react differently from the T cells, react differently from the cancer cells in that tumor sample. We can start to say which one of these compounds that we're designing is actually better for this patient population or is having the desired reaction that we want to go after. You may call that all translational science. Mm. Absolutely is. But your translational science should match up to your compound design, which should match up to your clinical development. Okay. All right. So now I guess with that all clarified, I think maybe let's touch on the partnerships. Yep. You know, I think it's, it's always critical at this stage to think of a company's development, not just finding good partners, but kind of laying out a strategy for why you would pair with a particular entity, what the focus is on that. And I guess maybe give us a sense for each of the three that you now have confirmed. I mean, beyond potential milestones and royalties down the road, I mean, why, why now with them? What do you guys get out of this for the partnership as it stands today? Yep. So I'll answer that from a couple of different perspectives. Let's start with the economics of it, because it's just simpler. We cannot possibly take forward a dozen compounds at a time, all on our own balance sheet. Mm. Even if we have the capacity to be able to do that from an operational perspective, we can't possibly do that from an expense perspective. And so by bringing in partners, we're actually able to utilize our capacity in a more efficient way. way. And so that was part of the original reason. The other part is, it actually makes us a lot better, because every time we do a project, our platform grows, learns, we get more data, we solve more problems that can potentially solve other problems, for future projects. And so that's been a really important aspect of it. Now, we've brought in $225 million from partnerships so far. That arguably has a dual purpose then. So we've been able to advance a lot of programs that'll be great economics if they become drugs as well, even before that. So for Sanofi, for example, each individual program has potential milestones of $343 million. $193 million of that is discovery or development based, so this isn't, you know, back-end focused. And then our royalties on those projects average in the low double digits. So really, really nice economics for that. But we're essentially having the partnership pay for all of the direct costs ahead of time to build out that pipeline, that long-term economic stream, and also that's building our platform and making us better at what we do. Okay. So from where you guys stand today with the three partnerships on board, where you're standing in clinical development, and we'll get to some of the pipeline programs coming up. Is it a priority now to continue to build out this BD portfolio? And I guess I'm really asking from, from a, from a platform perspective, is this something that you intend to become full-scale commercial entity with your own wholly owned assets, or do you kind of envision the platform itself as being kind of what, what brings you to market through the ventures of other partners? Sure. So on the last part of that, hopefully, you and this audience will respect that I'd like to keep our option value open as long as possible. So we don't have to make the decision on whether or not we're going to do late-stage development and commercial launch. But we have a rule internally, anything that we put into our internal pipeline, we have to have a pathway to be able to do it ourselves, so that we're not ever required by a part-- required to get a partnership to be able to advance it. Sure. So, I'm not sure if we'll ever commercialize our own drugs, but we should be able to, and that should be an option for us. Mm-hmm. Realistically, we'll have to start making decisions on that, probably in about a year, if we want to be able to do that in the future. We would only do that in oncology, which is where we're building all of our early development expertise. And then it would be a matter of, do we wanna set up the infrastructure for pivotal trials? That would be a lighter lift, but still you have to be able to manage a larger clinical network. The big question then comes in: Do you commercialize it yourself? Sure. So we are set up so that we could outlicense or divest assets without it meaning that we have to do the same for the entire portfolio. Okay. Are there any fundamental differences in the way that you're developing the platform itself, whether it's automation facilities or what have you, that would potentially be skewed one way or the other based on that kind of decision making or? Yeah, it's, it's a good question. We're actually right-sized for infrastructure right now. So we just opened up our automation plant, which was a big 3-year project. So, what our automation plant does is it basically allows us to, we do generative AI, modeling systems, active learning, all virtually, that gets us to compound selection. Right now, we can make some of those in-house, but a lot of those get sent out for testing. With the automation facility, we're actually able to feed it directly into an automated facility that will then chemically make the structures, and then in the same loop, be able to test them across individual biological assays, and feed that data directly then back into the system. So you can see how it's a fully integrated loop of virtual learning and then experimentation. So, that is just opening up. Now we've been running the biological assays, which are running phenomenally, and we've started to make the chemistries as well. That'll give us a lot of capacity and a lot of efficiency. So we don't need to make a lot of investments anytime in the near term to continue to work at the levels we are or even a bit more. We continue to find a lot of efficiency. We've actually brought down some of our R&D costs despite growing our pipeline- Mm. And that's all been efficiency work. Okay. I mean, that's a perfect segue into the pipeline itself, right? Yeah. I guess, high level, what are the important pivot points of the next, let's say, 18 months, that we need to be aware of? Yeah ... for the four programs that you've announced, either in or about to be in the clinic? Yeah, it's a fun time for us because the early supporters of the company were believers in transformative technology and what a platform can do. And we have always been followed by the biotech community, but unless someone had a lot of interest in the tech components, most of them have been more on the sidelines. And this year they're gonna start to get clinical data, which would be, we know, really interesting for them, and certainly what a traditional biotech investor would be used to. So CDK7, it's in dose escalation right now. We, as you would expect, run a very statistically efficient trial, so it's a adaptive, continuous trial that can transition directly into the phase II portion of it. So, we would expect that transition to happen in the second half of this year. That started the dose escalation monotherapy in July. And the enrollment's been going well. The first phase II that would kick off would be in monotherapy as well. This is a mechanism where you could see monotherapy efficacy because it is both a cytostatic and cytotoxic, so that's exciting. But then we will also start a combination dose escalation shortly after the monotherapy transitions into phase II because there's a lot of potential to use this in different combinations. So a lot of people are looking forward to that data. PKC-theta, that's the drug that BMS in-licensed, started its phase I testing in February of last year. We don't have specific updates besides that we know that it's still running, and BMS will hopefully make an announcement on where that's going sometime in the near future. You would expect it from a normal phase I timeline. And then, we're putting LSD-1 and LSD-1 into the clinic this year and giving more updates on MALT1. The important thing to recognize is actually a lot of our early pipeline is based on areas where it's a really attractive target, but there's some critical or multiple critical design flaws that no one's been able to solve before. And so most of the time, that is not potency. PKC-theta is a little bit different, but for all of our oncology drugs, for example, you know, people were able to come up with potent compounds before, but they weren't able to come up with compounds that were safe or could be dosed properly or didn't have other critical flaws that made it highly predictable that they wouldn't be successful. And so the reason this is important is the phase I data will actually tell you, are our design platforms working and accomplishing what you want them to? Because you'll be able to see that in the PK, the PD, the safety data, and say: Did they solve the critical problems in these targets, and are we actually, for the first time, going to get a good look at this biological mechanism? All right. I mean, that tees up my next question, actually, perfectly. I mean, when you're talking, let's take CDK7, for example. Yeah. You know, it's not the first attempt at targeting CDK, like you've mentioned. Yeah. I guess, give us a sense for this compound specifically, what are, what has your platform allowed you to do that gives you confidence you could potentially avoid some of those pitfalls? And then, I guess, from a strategic perspective, to your point, you could see some of those answers in a phase I. Yep. How are you kind of dual tracking next steps after you see that data, based on what it could look like left to right? Yeah, yeah, of course. So, CDK7 is very much a therapeutic window problem, and that's where basically everyone's fallen over. It's been on toxicities, right? Because it's sitting at a really critical point in a cell cycle, and you're inhibiting both cell cycle and transcription. Now, this is potentially really exciting because if you look at the CDK4/6s, you know, $9 billion drug class, the reason a lot of times they're they have PFS issues is because transcription is upregulated as a adaptive mechanism in a lot of those cancers. And so if you're able to inhibit both the cell cycle and the transcription, you have a much better chance at being able to suppress adaptive response. So CDK7, really attractive target. The critical point, though, is every cell needs to be able to go through a cell cycle. And so if you look at almost all of the drugs that came out in the CDK class originally, they were almost all covalent inhibitors. Now, for a medicinal chemist, putting a covalent inhibitor is a way to really boost potency, right? It's, it locks down your compound. And so a lot of the first-generation compounds you see come out are often covalent, and it's just sort of shorthand for increasing potency. But you absolutely don't want that here- Mm. Because you're going to be attaching irreversibly to a lot of things that you want to have normal cell cycle. And so you have to get in, hit, and get out of the body. And so when we were doing our workup, we looked at it at about a 6 to 8-hour window is the right period of time where you can get in, preferentially hit on the cancer, get out of the body before all of the rest of the cells have had a chance to do their normal cell cycle, and you're bringing in. And so that was the first thing, is we've got to get this dosing right. And that's actually something that we can plan for from the very beginning and say: I need it not only to be potent, not only hugely selective, because CDKs, you really don't want a pan-CDK inhibitor, you really want it to be selective. But also, I need to get into the body, hit my target, and get out. And so, we designed for all of that. Another thing we noticed, though, there's only two drugs that had reversible inhibitors. Both of them were substrate for transporters. This is gonna be a huge problem with something like CDK7, because what it means is you're gonna be recycled in the gut, because that's a transporter-rich environment, so it's gonna keep kicking that drug back around in the gut. Your gut's gonna have a lot of cell cycle going on, and so you're gonna have a ton of gut toxicities. Then you're gonna have variability on your PK because your absorption is gonna be all thrown off. And then, if you make it to the tumor microenvironment, it's a transporter-rich environment, and so staying in there to treat what you're there to treat is gonna be really hard. So, the reason I go into some of that detail is twofold. One, this is something that is normally never talked about in a biotech presentation, right? Like, there's not a nice mouse model graph that's gonna show you this. But in fact, we looked at that and said, these drugs are very likely to fail because you're not gonna be able to get the dose that you want, and you're not gonna be able to stay on target. And so that's how we profile a drug. The other really important thing to remember is clinical trials succeed or fail, not on biology, they succeed or fail based on statistics. And so most of the clinical trials that you look at, you have to factor in how many patients stayed in the trial, what are all of the different things that change your N inside of the treatment group that you're looking at? And so if you think about CDK7, if you've got gut tox, you're gonna have patients who walk out the door before they've even dosed more than a cycle or two because they're gonna say, "Hey, I've got grade three diarrhea, I'm leaving," right? You're also going to have huge PK issues that are gonna lead to some patients that are wildly underdosed, and some patients that get too much of a dose because there's that high variability. If you have variability in your PK, you have to push to higher dose to make sure that you've got an average across all of them. And that's also gonna be more side effects or less efficacy if you're not in the right range. More side effects means more patients leave. Less efficacy means those patients progress and leave the trial. Before we've even reached the tumor, you've already statistically weighted that trial against you. Mm. These are the things that people need to think about when they're doing drug development, and that is really, really hard to optimize for in a traditional drug discovery process, because in a traditional drug discovery process, you find hits based on potency. You find a few hits, and then you say, "I need to make that selective." And then once you've got a hit or two that looks potent and has some selectivity, you optimize for everything else. What our whole model is, is go through, think about everything that you want in a drug from the beginning and try and optimize it. Okay. All right, that's great. That's a lot to unpack. It's fantastic. I mean, when you're talking about all these modifications that are important to consider for STICKs on PKC for a second. Does that impact, and thinking also about the tolerability, does that impact any specific indication, any specific tumor types that you could go into? Obviously, tolerability is gonna be a bit different with each patient population, but with the profile of your drug, I guess, what are the important considerations, like when we see the phase I data, that could potentially tell us, "Okay, this would be amenable to this indication, but not this," and then how do you plan ahead for that? Yeah. So, the planning ahead is a great question because what we did is, this is a mechanism that could be applicable in most cancer types, right? Because it's just rapidly proliferating cells. And so, what we really wanted to do is drill down into which categories, which areas had the most need for it. And so, we used our, patient tissue platform that I was describing earlier as, one of the ways of looking at it, and what we actually saw was a divergence, two different cohorts, essentially. And one cohort was responding at literally an order of magnitude or two orders of magnitude lower dose than the other cohort. Digging into this, what we really found was there was a high-grade nature to this, which makes sense, because the more of this, the more a tumor becomes aggressive, the more likely you are to have that transcription upregulation and the more cell cycles you're going to have. But that also led us down a pathway where, okay, now we can actually define more what is the phenotype of this patient population we want to be treating? We can go after cancers that are more likely to have this phenotype then. Then that should weight the statistics towards us. This is where I said, we think of precision medicine in a little bit different way. Right. That wasn't meant to say, we're gonna target this 1,000-patient pool, but we're gonna go from 100% down to 80%, and that by cutting out that 20%, we're improving the statistical powering of our trial. Okay. So it's about knowing your patient. It's about having a better sense of where in that treatment paradigm the patients are gonna be that are most likely to do it. So we also, obviously went through a lot of literature, did a lot of biology research, and we picked out six indications that are currently running in our dose escalation. Okay. So phase I data coming up, and then potential phase II start. I think you said early next year, or shortly- Well, it's a- After phase I? a continuous adaptive trial. Right. Okay. Hopefully it'll start very shortly after the phase I. Okay. 'Cause it's a real-time data model. Okay. Now, then maybe into PKC-theta. When can we expect maybe the next update? And I guess, you know, maybe a little bit more focused than some of the dives into the prior compound, but- Yeah ... I mean, how are you thinking about that strategically? And it's kind of the same process, right? Yeah. Like, when you get the first data, what's the most important data points for you all? Yeah ... to understand whether or not it's worked, and then what your next step should be? So this is one where literally the mechanism's never been tried. More than a dozen companies have announced PKC-theta programs, many of them large pharma. None of them were able to come up with a compound that was both potent and selective. And so, we were able to design the first in, it was about 12 months from initiating design till we came to our development candidate. From a chemistry property, it looked beautiful. If the PK/PD work comes out well there, it literally will be the first time people are going out, or were going after this. This is a big potential I& I drug. You could use it either in combination or potentially in replacement of some of the biologics that are out there for some of the big markets. That's why so many people were interested in it. Mm. That'll be a little bit of a drop the mic moment for us if we were able to do it quickly and efficiently, where a large pharma really failed. That'll be a great validation for us. Okay, great. It is something you'd know from the PK/PD. All right. And I think just in the last minute or so here, I do want to ask for the LSD1 and the MALT1 programs, I guess timing to get into the clinic, and then I do just have to ask, I mean, is there an intention this year or next year, or do you have a sense of timing for any additional programs that you might announce? Yeah, of course. LSD1, really short, looking to start first in-patient for at least one indication this year. Most likely, there would be AML or small cell lung cancer. Great biological rationale for both of those, and we've got a really differentiated compound there as well. MALT1, we think the UGT1A1 inhibition is a real problem for the other MALT1 inhibitors that are out there, mostly because one of the biggest potential markets here is CLL, where you'll almost certainly be given along with BTK, which have its own liver tox issues. UGT1A1 increases your bilirubin levels. You don't want to have both of those going up at the same time. That's how you trigger Hy's Law. And so, we'll give another update on development pathway for that, too, for second this year. Okay. And then, yes, we have more things coming. All right. And on that note, I think we are just a little bit over time, so I wanted to say thank you, everybody, for sticking around. It was great. Thank you for joining. My pleasure. Thank you.