Hello, everybody, and welcome to our Goldman Sachs Healthcare Conference. My name is Chris Shibutani. I'm a member of the equity research team and here, together with my colleagues, including Karishma Raghuram. We are very excited to have Exscientia here join us once again. This year we have Ben Taylor, Chief Financial Officer, but really Chief Wear-So-Many-Hats Officer, and able to pull that off. Tell us a little bit about you, including your deep, dark, secret past of being a former GS alum. I was, yeah. I spent about a decade at Goldman, actually. So, 15 years in healthcare investment banking, all of it biopharma focused, but have now been on the executive side for about 10 years in a mixture of operational and finance roles. Yep. You know, you voted with your feet about where to go. Yes. You made some prescient decisions because certainly anybody who's going in the direction of the future, undeniably artificial intelligence is just part of that vernacular there. What was your thought process and the risks that you were taking, and how have things been playing out? Because, like any true pioneering adventure, nothing is linear. Yeah, no. Well said. So, what got me to move out of banking was actually, I saw so many companies developing drugs, and you could see the business models around it, you could see the investment choices, but you weren't seeing a lot of difference in the actual patient outcomes. Our bars for success are actually so low. Like, we think about a 30% success rate in a clinical trial as being, you know, phenomenal and approvable, but that means that 70% of the patients in it fail. Correct. And so, that means that you've got a culture where there is less risk-taking, there's more focus on trying to do what is immediately achievable, and there's sort of a dislocation between the science and the fundamental investment models. And so, what I wanted to do was step out and not just step into a traditional role, but really go for how do you try and tackle transformational change in the industry to change that probability of success. Because the fundamental problem that we all face, whether it's as an investor or as a manager, is if you are in a 95% failure environment, which is what biopharma is, you can't make rational decisions. You need to have more data, better models, better ways of informing yourself. And so, when I spent several years at a biotech running the day-to-day operations, but then when I found out about Exscientia, it was basically doing exactly what I wanted to do, which is how do you break down the problems that are in clinical trials? Because this is another common misconception in the industry. People think of a clinical trial failing as being related to the biological mechanism, but it's not. Clinical trials fail for statistics. And so, what we want to do is be able to break down all of those reasons that a clinical trial might statistically fail. That can be drug design, it can be mechanism, it can be patient selection, it can be trial design, and figure out better solutions for that. At Exscientia, what we're doing is really focusing on the drug design portion of it and trying to take it out piece by piece and saying what might cause a phase I failure or a phase II failure in this particular drug for this patient population and design a better drug so that doesn't happen. Yeah, no, it's interesting. There's always this concept right now of trying to figure out how to measure some of the benefits. As a public company now, with this ecosystem of the investors who are in the room, we're all kind of like hardcore therapeutic biotechs. I don't care whether it was some sloppy process done by some fill-in-the-blank large cap that they spent a lot of money they probably didn't have to do and came up with a molecule that's got a huge zit on the corner of its head and is going to take it into the clinic and all this sort of stuff. And the fundamental premise of a lot of the analytical process that you're going through is to be more specific, better, faster, more efficient, cheaper. Therefore, there should be some reward of being able to sort of say, okay, this molecule now officially tapping down goes into the clinic and it's probably a stronger asset. The public equity investor has a really hard time measuring that. So, how can we get confident that process is? I sort of feel like a lot of the AI R&D companies are at this juncture where they're like putting their first kids into the clinic or whatever and we'll measure you from there. Is that right? Or is there any other way that we can perhaps be willing to perhaps appreciate some of the preclinical stuff? Because you're never going to get help from the big pharmas because it's invisible, all this stuff. We know that there should be a lot of failing and it's expensive, but your processes are probably smarter, faster, cheaper, more efficient, et cetera, but you don't get credit. How do we solve for that dilemma to get credit in the stock? It's a great question. So, I think what we've tried to do is lay out a roadmap. And so, you've been following us for several years. You'll remember there have been a number of programs, four or five, where we looked at it and we said, this program is going to fail and this is why. And you can see it in the chemistry. And so, I think we have a tendency in biopharma to look at potency, right, and say, hey, this is making a tumor shrink in the xenograft model, which is really a measure of potency largely, and base a drug's potential success on it. But if you actually look at the properties of chemistry, you can start to say, this is going to fail because of off-target issues or absorption or different things like that. And so, several times now we've stuck our neck out, we've called out that drugs are going to fail a clinical trial, and so far we've been right every time. And so, I think that was a message where you can say, okay, you can actually predict these things earlier on. You don't have to spend $50 million to run the clinical trial before you can figure it out. You can actually figure it out from some lab tests. Then what we tried to do is lay out, and we're very public about our profile of the chemistry going down much deeper than just potency and selectivity. You really want to look across the properties because then you can start to see how is this actually going to interact with the system. This is something very quantifiable because remember, even though we use AI, to your point, it kind of doesn't matter because we produce a chemistry and that chemistry can be taken into a lab and quantitatively tested for properties. So, you have a quantitative basis for looking across a group of different potential compounds and saying, this one's actually better. It has more preferential properties that you would want. Now, that starts to translate then when you think about how does this play out into the clinical trials. Phase I has about a 50% failure rate. Most of those are actually going to be design-related. It's about how does the drug get to where it needs to go? How does it stay there? What are the interactions? Because the biggest problem people have with design is they think if I can drug a target, I've got a compound. Well, remember, when you give a person a drug, it's a systemic effect. So, you're actually interacting with whatever, the 20,000 different proteins that are circulating through the body, the different genes. And so, you actually have to be able to predict it on a more broad basis. So, when it fails in phase I, that's almost always going to be a design issue, not a biology issue. And probably about half of the phase II failures, which is another 70% of drugs failing in phase II, are going to be design-related. So, actually, in the very near term, you're going to get a look because remember, we've got six drugs that are either in clinical trials or about to go into clinical trials. So, we've got one that's in phase I, II testing that you get the clinical data on the dose escalation later on this year. BMS, we just had a successful initial result coming out of our PKC- theta inhibitor that's in-licensed by BMS. That's obviously hugely positive because this is a target where over a dozen companies spent hundreds of millions of dollars trying to design one. All of them failed. One of them actually came to us and asked us to design something for them. We did it in less than 12 months going from idea to identifying the development candidate, and it's now the first potent and selective inhibitor for PKC- theta. So, that was a nice proof point that already came across. And then we've got two more that are partnered with Sumitomo that are in the clinic. We could get first data on this year and two more of our own that are going into the clinic around the end of the year. Undeniable enthusiasm for talking about the proprietary pipeline, but I think the brainiacs on my team, including Karishma, who has her degree in stats, loved your answer about the distinguishing characteristic about the platform being statistical in nature. So, did we cover that appropriately enough? Was there a secondary question that you think I should dig further into? Are we good? You're smiling. Hi, from the audience. So, I guess my question really is, because your assets are mostly mainly in oncology, are you worried that perhaps the parameters that you're testing your model on are not going to be agnostic enough to then be able to be applied to other therapeutic verticals perhaps because of that high level of specificity? You might be overfitting, and that could be a big point of, I think, contention. Yeah, outstanding question. So, we actually think about it as an axis. If you think about our design platform, it's therapeutic area agnostic. All we need to know is a good looks like, and we will design a molecule for what is the baseline that you're asking. Think of it as like. So, that means you have to know what good. If you have a way to measure that, you have to know a way to measure a variable depending on target, depending on therapeutic area, depending on what the output you want. That's why we only design drugs in oncology, because if you think about our design platform as being primarily chemistry-based, the biology is highly specialized across different therapeutic areas. So, we have developed our own internal biology expertise to be able to develop oncology systems. Now, when we work with Sanofi, almost all the drugs we design are immunology and inflammation, but they're a world leader in immunology and inflammation. They know what a good drug looks like for it. And so, we can rely on them to define what good looks like, and we can just design. And so, we always are looking at both and trying to figure out, do we have a good translational model, essentially is what it comes down to? Because if we do, then we can design a good drug. Love the audience participation. Yeah, this is great. Very compelling. Let's talk about your home cooking, the proprietary pipeline, a couple of different assets, journey into the clinic. Number one, 617, the CDK7 inhibitor. Tell us about this asset, you know, opportunity positioning, CDK7 inhibitors. You know, there's an interesting backdrop here. CDK4/6 has certainly earned its stripes commercially, you know, gotten three big bad boys involved with competing with each other, very significant revenues, and yet there's always going to be unmet needs and opportunities there. How does 617 potentially fit? Yeah, great question. So, we like CDK7 as a target quite a bit because if you think about it, it's actually got a dual mechanism. You have both the CDK, the same that's going along with the CDK4/6, which is with cell cycle inhibition. CDK4/6 does it at an earlier phase of cell cycle, but it's similar in mechanism. But then CDK7 also has transcription inhibition. Now, that's really important because if we look at the CDK4/6 inhibitors, a lot of the adaptive response that we've seen has been upregulation of transcription. So, then hopefully you're preventing some of the adaptive response that we've seen out of the tumors. The other thing, there is a broader mechanism to some extent with CDK7 because it actually does affect CDK1, 2, 4, and 6 ability in it than some other drugs. But what is, because you're at that more centralized position, toxicity is a potential issue. And so, what you want to do is really carefully control your CDK. And so, when we were designing it, most of the CDK7 inhibitors that are designed, first of all, were irreversible, which we don't think is feasible for this because you're just going to have a lot of fundamental shocks across the system. You need to be able to get in and get out. And in fact, in our preclinical models, what we've shown is about eight hours is the predicted amount of maximum effect on tumor. Any more than that, you're actually not having an effect on the tumor, you're just increasing toxicity. And so, what we wanted to design it for is not only to be reversible, but also to be able to get into the system and get out in about six to eight hours. And so, those are some really key components about it. Now, there only have been a couple reversible compounds that have been designed. Both of them, interestingly, had a very little known issue as transporter substrate. So, this has probably never shown up in any biotech presentation that you've seen, but it's actually fundamental here. So, in your GI tract, you have lots of transporters that basically will kick things out of the system. If you're a transporter substrate, what that's going to mean for a drug is you're recycled in your GI tract. If you're a cell cycle inhibitor getting recycled in your GI tract, it means you're going to be disrupting all of your GI cells. You're going to have a lot of GI tox. You're also going to have a lot of absorption problems. Now, going back to the statistics, for a clinical trial to fail, if you have a patient tox out because they've got grade three, grade four diarrhea and they discontinue, that's the exact same thing as not having a biological mechanism that works because you've lost statistical power for your trial. So, we knew we had to design around that to make sure we could get good absorption. But then, ironically also, if you do make it into systemic absorption on the level that you want and you do make it to the tumor microenvironment, the tumor microenvironment is designed to pump drugs out of the system and has a very robust transporter system. So, being able to stay in that tumor means you're going to get pumped out if you're a transporter substrate. Now, that's a long convoluted explanation, but it's also really important because these are not things that we talk about in drug design. We talk about potency. We don't talk about efflux. But that drug, those drugs, one of them has been discontinued, we predicted would fail because of not potency or selectivity. Actually, the potency and selectivity on the one that was discontinued looks great, but you're never going to be able to get where you need to go and stay where you need to go if you've got this transporter issue. And so, that's why people need to take a much closer look at the full design and where we can make a difference with our platform in being able to just design a more balanced drug. Let's take names, not prisoners. The discontinuation drug, was that a Lilly asset? Was that a Carrick? Carrick's still running, I believe. Okay. So, Syros and Carrick were the ones with the reversible mechanisms. Lilly and Exelixis had irreversible mechanisms, which have, I believe, also been discontinued or downplayed at least. Slow walk. Yeah, no, exactly. It's often with these novel targets that we see some validation when some of the other players are going in, and it becomes an opportunity for investors who are paying more broadly attention to pay attention. So, tell us about the clinical trial, ELUCIDATE. Tell us a little bit about design, basic building blocks, numbers of patients, endpoints. Sure. That we're going to learn. Sure, sure. So, adaptive, expandable phase I, II trial. We're going through the monotherapy dose escalation right now, which we'll read out later on this year. Progressing well, enrolling well. What that would do is lead into then a combination dose escalation. Should be pretty short. So, probably three to six months additional on the, you could call it a phase I-B with the combination dose escalation. I'll talk about the combinations in a second. And then that would be able to seamlessly go into an expansion on that indication. So, phase II, basically. As far as patient numbers, the phase I is designed to go up to effectively about 30 patients, but it's a CRM model. So, it's basically doing continuous reassessment to reach the optimal dose. And so, it could be a different number than that, but it's doing real-time assessment of the data to predict the optimal. Typically, we think about breast cancer, but are there tumor types that you're contemplating? Yeah. So, breast cancer, definitely. If we look at a clear starting place, whether it's Pfizer's CDK4 or Carrick's initial data on the CDK7 in positive, HER2- CDK4/6 refractory breast cancer, they've both shown signals. So, the fact that a CDK4 can still have signal in CDK4/6 refractory, very positive, obviously much more opportunity for a CDK7 to have impact there. So, that's a good starting place and likely where we'll do the first combination studies in the phase II. But something really interesting, if you look at the preclinical models, there's actually a very strong rationale to combine with immunotherapy. The effect of the two appears to be independent and cumulative. But none of the clinical trials that have been tried in those spaces have been successful. The reason being, they're basically different side effect models. The immunotherapy side effects and the CDK side effects tend to be in different categories. If you put them both together, you get a double effect. Now, if we are able to manage the side effect profile, the therapeutic window better, that actually opens up the co-administration with immunotherapy and very large market opportunities that aren't available to other ones that wouldn't have that same therapeutic window. You might think of lung cancer, either on the small cell or the non-small cell side. Both have shown really, really exciting preclinical data. That's some place you might see us go as well. Just remind me, I just want to remember to ask here about the geographic footprint. If we are conducting your clinical trials, the supercomputers in Oxford, you guys have kind of a global footprint. You know, the chief medical officer has been very familiar in Europe. We just came from ASCO, where there was a lot of very exciting data that came from China. What is your strategy in thinking about where you want to make sure you're gathering data? Yeah, so focus on the U.S. and Europe. Now, where we start actually depends a little bit on the program. For lots of good regulatory reasons and enrollment reasons, sometimes it's better to do in Europe, sometimes it's better to do in the U.S. We're sort of ambivalent for the phase I. For the phase II, we would always want to do both. Okay, got it. Let's move on to another asset, EXS539, LSD1 inhibitor. Yep. About to go into the clinic. I think back into last year we submitted the IND. Where are we with progress? I think we're actually potentially going to get some insight into this over the course of this year. Yeah, so should start in the clinic later on this year. Yep, but should go in shortly. Okay. Looking at AML, the differentiation here, so LSD1 is a really interesting mechanism, basically on differentiation for cells that could have a lot of impact on things like AML or small cell lung cancer. And there's been some encouraging data coming off of LSD1 inhibitors in AML already. The problem is it's very closely related to platelet production and blood cells in general. So, you really have to watch the hematological tox with it. Okay. And this again, sort of if you don't have the PK profile right, if you use an irreversible inhibitor, you're just going to have a really high level of hematological tox, which is going to limit your dosing, which is going to limit your efficacy. And this is what we've seen in a number of the indications that are out there. It's interesting, one of the LSD compounds that it's probably the best one that's out there right now, they're actually using it in essential thrombocythemia, so elevated platelet levels. So, you're basically using the side effect to treat the disease, but that's going to be really hard to carry over into an oncology setting where you already have this highly depressed hematological patient population. And so, this is again where design just comes in and is so important in being able to unlock the potential of that mechanism. Got it. Competitors who are in this realm? It's been a mixed track record there, but to name some folks there, is there anybody that we should also be paying attention to? I mean, Horizon is probably the most advanced in the space right now, but again, they're the ones going after ET. They are also testing in AML. There have been some actually really encouraging AML results, and that'll probably be the first place that we go to first as well. The issue is just they come with a lot of heme tox. And so, if you know 80% of your patients are getting grade three, grade four heme tox, you're going to be limiting yourself on adoption. The other thing that we designed into LSD1 is it's actually brain penetrant. And there's a very odd thing about LSD1 as a target where you've got three points that counter each other: the potency, cardiovascular risk, and brain penetration. And so, most people have solved for cardiovascular risk and potency, but that means almost nothing is brain penetrant. This is actually really important if you want to go into small cell lung cancer because about 50% of small cell lung cancer patients end up with brain tumors. And actually, the leading mortality cause in lung cancer is brain tumors, but almost all of the treatments for it don't penetrate the brain. And so, we wanted to design something that would be able to treat those patients as well. I think the third that's about to get primed into the clinic of your own, wholly owned, would be the MALT1 565. Tell us about that. I think we're talking about an IND in the back half of this year. Yep. Yeah, LSD1 and MALT1 are pretty much on the same track. And this is where it is most useful in B-cell malignancies. Now, there are certain indications where you could use it in monotherapy, but they're pretty small indications. So, DLBCL, ABC subtype is the main one where you would use it in monotherapy, potential mantle cell as well. However, when we were designing it, we really wanted to go after the core heart of the market, which is CLL. Problem with that is CLL is almost certainly going to be co-administered with a BTK or maybe a few other drugs that are similar, but probably a BTK. And BTK comes with liver tox. Now, when we were looking across the class of drugs, what we saw is some of them have nice potency and selectivity, but all of them seem to have a UGT1A1 issue. UGT1A1 is involved in basically the production breakdown of bilirubin. Bilirubin. Right. And so, if you have both elevated bilirubin levels and elevated liver enzymes, you're going to be triggering Hy's Law, which is a common thing that the FDA looks at because it's been shown to significantly increase mortality. Exactly. And so, we didn't see that as being viable for co-administration on the most important market for MALT1. And so, we designed that out and basically went down a just different chemical space completely to design that out. So, really beautiful looking compound. And we'll start that off in B-cell malignancies broadly as we do the dose escalation and then focus in on CLL. Great. So, you'll have three into the clinic into the holidays here, and then we'll be bugging you about when we're going to get those cards to turn over in 2025. Exactly. You did bring up some of the partnership here and particularly the profile with the Bristol partnership. We had Chris Brenner on stage earlier, and there were so many things that they could talk about. Obviously, PKC- theta has been a bridge too far for many folks for a long time, but we're sort of seeing some evidence of breakthrough. What have we learned recently, and how should we think about this? Yeah, really excited for where that compound can go. And this is a great example of where this target category. So, PKC- theta, fundamental immunological mechanism, but the family of PKCs has a lot of impact across your immune system. And it's a really difficult target to drug. So, we joke that we gave the rest of the industry a 15-year head start. Most of the large pharma, I think all of the large pharma that are involved in immunology gave it a try. Hundreds of millions of dollars were spent. Only two or three programs even made it to the clinic. None of them were able to get through for design issues. Two of the ones that went in were pan-PKC. And so, you saw a lot of side effects come out of that and really hard to actually get to PKC -theta as a target. And one of them was very selective, but absolutely terrible potency. And actually, the maker of that compound came to us, this was Celgene back in the day, and said, "Can you help us?" And within about 12 months, we were able to go from start to finish on getting to development candidate. And it's potent, selective, nice, clean, balanced profile. And so, that looks like it could be not only best in class, but first in class. And that's sort of how we think about a lot of our drugs. First in class is about getting across the line, not just about starting a design program. Right. Okay. I sequenced our discussion across these assets to create a glide path to an important part of your business model, which is the fact that separate from your home cooking that you're planning to continue to nurture and develop your proprietary pipeline, but there is foundationally and historically within the company this whole partnership model. It's been a very important source of really helping capitalize the company, keep the lights on, et cetera. Talk about, obviously, and Bristol represents the most recent shining example of that, but talk about what that means in terms of your strategy going forward, doing more of these, have they gone well, are they exhausting, you know, what are the emotions and what should we think about for the future? If you're not exhausted, you're probably not doing something interesting. Fair. But yeah, the partnerships, I think obviously it is and will continue to be a core part of our business. We've got major partnerships with Sanofi, Merck KGaA, BMS. And you know, we're seeing great progress. The Sanofi partnership has been off to the races and really terrific. And in fact, we took one of the compounds that we were internally developing and moved it into the partnership because just a really high potential product, but we just felt like they were a great partner and would develop it well after we do the design work. So, we continue to see a lot to come out of that. We've already brought in $230 million in non-dilutive financing from those partnership activities. But if you think about the economics for those programs, they're actually terrific. So, even though we are limited to basically just doing the design aspects of it, we take it from idea and deliver it back when it's ready to go into IND. For Sanofi, the potential payments around a single program are $343 million. $193 million of that is pre-commercial. So, this isn't some big BioBox deal. This is real money early on, and it's pretty evenly spread across the discovery and development. But then on the back end, we also have really high royalties. So, you're talking on average, the royalty should work out to be low double digits. And so, that's a really nice economic package, especially because we're basically paid in advance for all of the work that we do on it. So, yeah, I love that model. And it basically utilizes the extra capacity that we were talking about earlier. So, we couldn't possibly internally develop all of the compounds that we can design because the cost would just be too much. It would be across too many therapeutic areas. So, what makes sense is we choose one or two a year that we want to do internally focused on oncology. All of this excess capacity, because we are a tech-based platform, then we can utilize through partnerships, and we plan to continue doing that. We actually are likely to get another step change in productivity from what we're doing on automation. Right. This might be a good transition into that. Absolutely. But we just opened up our automation plant in full now. So, we've been running basically what we did is seven years ago, we realized you need to integrate experimentation with AI. That's the only way you can have tight learning cycles because we haven't talked about it today, but we don't see ourselves as a screening company. It's a learning company. Our potential space of chemistries is just too far to screen. And so, how do you learn quickly to get to the best result? And so, to do that, you need to actually control the data. Almost all of the data that we use is proprietary generated, but it has to be very specific to answer your question. You can't create an LLM to search the noisy data that's out there when you're trying to make sure that you place this atom in the right place on this compound, right? You actually need very specific question and answer models. And so, seven years ago, we started to bring in inside a lot of the what we would call assay development, basically the testing systems. So, pharmacology and all of the protein sciences and fragment analysis and different pieces, but we were still working inside of CRO's ability to make compounds and test compounds. And so, about three years ago, we started to build an automation facility. And now what we can do is basically directly from the feed of our AI design systems, do chemical synthesis, purify, test on biological assays, and then feed the data back into the system. We've got enough data on the biological assay system side because that's been running for about nine months. I mean, we're seeing a 75%-90% improvement in productivity on both a cost and time basis. Things that were taking months take weeks, weeks take days. That is not only transformational in terms of our cost and time, but also on your ability to do science. If you've got a six- to eight-week cycle where you're waiting for someone else, you can only ask so many questions inside of that cycle. You have to batch it. If we can say, "I just want to make one or two to ask a specific question, and I'll get the answer back in days or weeks," we can actually ask a lot more questions. We can do a lot more interesting ways of doing hit finding without needing to do a high-throughput screen, for example. There's all sorts of different variations we can go. So, it's a really exciting part. I think it'll integrate into our partnership business as well and allows us to play around with some different business models that could be really exciting. Let's suspend reality for a moment and just ask you a really narrow CFO-type question. Yeah. What would be cash runway and how you're thinking about being able to support all these projects? Yep. So, over $400 million in the bank, cash runway well into 2027. That gets us through, without providing any specific guidance, multiple milestones on at least all of the clinical programs that we talked about, as well as should provide validation across potential income generation for the business, the partnership business as well. And then for the transcript, we certainly wish that Dave, who is under the weather, is going to be feeling better soon. So, no conspiracy theories. But Ben, certainly you are very capable, wear many capes and hats, et cetera. So, a pleasure always to catch up with you. Yeah. Thank you very much for joining us this week. Great talking to you. Great. Thank you.