If you have any questions, please reach out to your Morgan Stanley sales representative. For this session, we have Relay Therapeutics. We welcome Sanjiv Patel, President & CEO; Don Bergstrom, President, R&D; and Peter Rahmer, Chief Corporate Development Officer. Welcome and thank you for your time today, gentlemen. Maybe just to kick things off, Sanjiv, could you give a bit of an overview of Relay? That will contextualize the questions that we have. Thank you for your time. No, thank you for the invitation. We're thrilled to be here. Relay Therapeutics, for those of you that are not familiar, we're a company that's based in Boston, Massachusetts. We're very well capitalized, and we have cash now into 2029. We're heads down executing across a variety of key initiatives. First and foremost, that is executing with our first-in-class, PI3K mutant-selective inhibitor RLY-2608. We've initiated a phase 3 trial in post-CDK4/6 treated metastatic breast cancer patients. That trial is up and running, and we're excited with that going because we think this is a very large patient population that's currently unserved. The second of our initiatives is that we are continuing to dose triplet combinations with RLY-2608 and fulvestrant, both with ribociclib and temaciclib, as we look to generate what could be a triplet combination that we can use in CDK4/6 naive metastatic breast cancer patients. Our final initiative that we push forward in the clinic at the moment is using RLY-2608 to dose a condition called vascular malformations that are driven by PI3K alpha mutations. This is a very large, unserved population, and we are very excited about generating clinical data in this area. Behind that, we have two preclinical programs that we've named: an NRAS-specific inhibitor, as well as a Fabbraze program, and we push those towards IND. We have a small research footprint now working on very hard-to-drug targets. We have a very nice portfolio in front of us, but really at the moment, it's all about heads-down execution as we start to generate clinical data that will really meaningfully generate value for shareholders. Wonderful, Sanjiv. Thank you for those remarks. We've got a couple of sort of macro type questions we'll ask in all our company, and hopefully we have a nice consolidated industry view at the end. With China's rise in biotech innovation, how are you thinking about your competitive position here? Will this influence your R&D and business development strategy? It absolutely does. Obviously, we see the very impressive rise of innovative work done in China, very specifically in the area that we're in, small molecules. We see now a whole host of licensing deals from the large-cap pharma with such Chinese-based, innovative biotech. It absolutely does change the way we think about things. We really only focus on first-in-class, very hard-to-drug targets, and that's our entire portfolio. We think the kind of days of doing kind of second, third, me toos are over because obviously you just cannot compete with the speed and efficiency of some of these companies. Yes, very impressive, what's going on in China and does actually very much change how we think about things. Wonderful. Thank you. As an AI-tech and now biotech company, can you describe the key ways your platform is leveraging AI and think about AI's future disruptive potential? Yeah, absolutely. Happy to talk about this at great length. Maybe I'll just give you a very concise answer, which is the rise of AI again. We're seeing it in all aspects of our lives, from driverless cars to how Netflix provides recommendations for us as we turn our apps on. I think in drug discovery, it's just a very hard process from beginning to end. It's thousands of very tough incremental steps as you go right through the process, from choosing a target to filing an IND and then going towards the NDA. For us, we see it as a tool that can help you increase efficiency and help you design the right experiments. We don't see it as a panacea for all, and we don't think any time soon will a drug be designed entirely by AI. We see it as an enabling tool, and we think that people will be involved in drug discovery for many years to come. Thank you. Last question before we dig into some of your programs. What has been the most impactful on Relay from the regulatory side? Has it been FDA, MFNs probably not quite relevant yet, or tariffs? Again, just given the stage that we're at, most of our regulatory interactions were done last year, as we set up our phase 3. As we come into this kind of new regime, we haven't really had that much interaction. We can't really comment too much on whether it's better or worse. In terms of being a preclinical company, we haven't really had that much impact on the tariffs and the various pricing changes that we've seen. We imagine that will impact us over time. On that one, we really haven't seen any change. Wonderful. Thank you. Now to ask about some of your programs. RLY-2608, allosteric mechanism of action, how does it differ from traditional PI3K alpha inhibitors and why is this significant for cancer treatment? PI3K is the most commonly mutated kinase in cancer. Over the last three or four decades, it's been at the center of drug discovery efforts across most of the large pharmaceutical companies. We've seen various iterations of this. This final iteration that we've seen over the last decade is non-selective PI3K alpha inhibitors. We know that this is a mutation that's implicated in a broad range of cancers. The challenge is, if you inhibit both the wild type and the mutant, you get a range of side effects, including hyperglycemia, rash, diarrhea, stomatitis. All of these lead to patients just not being able to stay on drug and leads to lower dose intensity and then lower efficacy. Unfortunately, we've just not been able to produce a drug that has been efficacious in this field. When we set out back in 2017 to try and drug this target, what we were looking to do was try and create an exquisitely mutant selective inhibitor. The premise was if you could dial out these wild type toxicities and only really inhibit the mutant, could you generate a better safety profile? In turn, could you keep patients on drug at greater dose intensity? Then could you generate greater efficacy? That's exactly the premise of RLY-2608. We were able to discover a novel allosteric binding site that was preferentially seen in the mutant, and that gives us a handle on mutant selectivity. Obviously, now we've shown a range of clinical data that starts to prove out the exact hypothesis, which is the data we show in RLY-2608 shows the potential for a better safety profile. That has now translated in the data that we've seen in the early trials into what we believe to be better efficacy. Wonderful. What clinical efficacy data has been reported for RLY-2608 in combination with fulvestrant? I'll hand that over to Don Bergstrom. Yeah, so we've reported data out of our first in human trial, the Rediscover trial. Our most recent disclosure was in June of this year at the ASCO meeting in Chicago. What we've shown in the most recent update is we now have our efficacy evaluable subset of 52 patients. These have all been treated at a 600 milligram dose of RLY-2608 acid, which gives us the exposure level that we'll take into our phase 3 trial. We see a confirmed objective response rate in these 52 patients of 39%. We see a median progression-free survival of 10.3 months. These numbers are significant because they're almost double what we see for other PI3K pathway targeting agents in analogous patient populations. Our comparator in our phase 3 trial will be Capivasertib. That was tested in the Capitello 291 trial that led to its approval in PI3K pathway mutated patients. The reported PFS in PI3K mutated patients in that trial was 5.5 months. We're sitting at 10.3 months. If you take into account the Capivasertib population, it was largely a second-line population. They had never seen any prior Fulvestrant. Our patient population includes a number of third-line and later patients where half the patients have seen prior Fulvestrant. If we go a little bit further and just look at our true second-line patients in the Rediscover patient population, we're sitting at 11 months PFS. Really almost double what we're seeing for our comparator. Amazing. Thank you. Could you please throw a little bit of meat on the bones around the safety profile, particularly regarding hyperglycemia, rash, and diarrhea? Yeah, we see very low rates of high-grade hyperglycemia. Where physicians really start being concerned with the management of hyperglycemia is when they make it to grade three or more. That's really when physicians need to start using injectable agents like insulin to be able to control hyperglycemia. This is where they're calling in endocrinologists to help manage the patients. In many cases, especially in busy community practices where there may be limited access to endocrinology consults, that's a point where the therapy is just abandoned unless the patient is really benefiting significantly from it. What we've seen for other agents in this space in typical Western patient populations is that grade three or higher hyperglycemia is typically in 20% or higher of patients, and we are in the low single digits. Wonderful. For the other TAEs, which are very taxing for patients, you know, diarrhea, rash, stomatitis, those can range. For some of these agents, you know, Capivasertib, almost 80% of patients have diarrhea. You know, you've got high rates of stomatitis. Again, for us, for high rates of any of these AEs, we're either in single digits or zero. Thank you. What is the current status of the Rediscover 2 phase 3 trial, and what are its primary endpoints? Yeah, so we're currently enrolling that trial. We opened it earlier this year. It's a 540 patient trial randomizing fulvestrant plus RLY-2608 versus fulvestrant plus Capivasertib in PI3K mutated patients who have seen prior CDK4/6 inhibitor therapy. Primary endpoint is progression-free survival. The way we've designed this trial statistically is to look at this in a hierarchical analysis. In PI3K mutations, there are two main classes of mutations depending on the specific part of the protein that's mutated. The kinase domain gives rise to what we call kinase domain mutations. Typically, these are H1047, is the specific amino acid that's mutated. Non-kinase domain mutations most frequently are in the so-called helical domain, E542, E545, are the amino acids that are mutated. We'll look first at PFS in kinase domain patients. We anticipate that will represent about half of our clinical trial population. Then we'll look at PFS across the entire patient population, so all PI3K mutated patients. We've really done this primarily because when we were designing this trial, there were, and there still are, a number of competitors out there who are actually designing drugs that just target these kinase domain mutant patients. Lilly had a drug that just targeted these kinase domain mutant patients, but they just discontinued at the very end of last year. What we've seen in these kinase domain patients is a 67% response rate and an 18-month PFS in our heavily pretreated patients. We're seeing very strong efficacy in these patients. In the anticipation of there being potential competitors out there at some point that just target that patient population, we want to make sure that we're able to have claims in our label with appropriate statistical design to be able to call out these very high response rates and high PFS. We've designed this hierarchical analysis for PFS. We also have OS as a key secondary endpoint, but the precedent in this space is for PFS to be the registration endpoint. Wonderful. Thank you. Could you map out the market opportunity for us for 2608 and how many patients could potentially benefit? I think people, maybe we can see that. Yeah, sure. As you think about the post-CDK4/6 metastatic breast cancer setting in the U.S. alone, there's probably about 13,000 PI3K alpha mutated patients there. That probably goes to about 30,000 in the major global markets. As you expand that into the frontline, and it also includes the CDK4/6 naive setting, you know, that starts to about double those numbers on a global basis. These are, you know, even the second, the post-CDK4/6 setting alone, it's a multi-billion dollar TAM that it would be an opportunity for us to start to penetrate over time. Wonderful. Thank you. How was RLY-2608 discovered using Relay Therapeutics' platform, and what structural insights led to its development? Yeah, so our research team, when we started working on PI3K, there actually had never been a full-length structure of the protein that had been solved. That is largely, you know, this is now largely a technical issue. It's the art of being able to actually make the protein and image it. We have very persistent and talented protein scientists who are able to make the full-length PI3K alpha protein, use emerging structural biology techniques like cryo-electron microscopy to actually be able to generate the first known full-length structure of PI3K alpha. We are able to take that information that we get from our structural experiments and then use our computational methods to be able to start simulating the dynamics of how mutant and wild-type PI3K alpha move and to try to identify differences between the mutated state of the protein and the non-mutated state of the protein. We started to appreciate there were potential druggable pockets that were opening up in the mutated protein that weren't readily seen in the wild-type protein. These pockets had never been seen in any of the publicly available structures up to this point. That informed our screening strategy to actually start identifying small molecules that bound to PI3K in those pockets that appeared to just be present in the mutated protein. We found a hit that eventually led to what became 2608. As we progressed this through drug discovery, we were able to show the mutant selectivity. We were able to show that it was potent against both kinase and non-kinase mutations, and ultimately able to optimize the profile of 2608 to give us the compound that we thought would give us superior efficacy in patients while avoiding the wild-type toxicities, which is exactly the profile that we've generated in the clinics. Would it be fair to say that AI drug molecule, this would be probably one of the furthest ones in terms of clinical development that there are globally? Yeah, I mean, I think we definitely, as Sanjiv mentioned, this was an AI-enabled project. It is a tool that we use across the various steps in drug discovery to make predictions that we can then go to the clinic or go into the lab and validate experimentally and then go back and iterate and update our models. Coming back to Sanjiv's previous point, this was not, we pushed a computer button and it spit out the structure of 2608. There was a lot of, I think, human insight that went into discovery of 2608 that then was supplemented by using AI and ML to be able to broaden the chemical space that we could look at, make new predictions about what might work, and then ultimately be able to go and validate those predictions experimentally. Thank you. Are you exploring or what triplet combinations are you exploring with RLY-2608? We've initiated triplet combinations with ribociclib, Novartis's CDK4/6 inhibitor, and then with temaciclib, which is an investigational agent that's currently being developed by Pfizer that is a selective inhibitor of CDK4. The idea there is selective CDK4 inhibition will avoid some of the hematopoietic toxicity associated with CDK6, give you the ability to get more effective CDK4 inhibition, and ultimately be more combinable. We need to remember that as we go into frontline patients, these are patients who are likely going to be on our regimen for two years or longer. Tolerability matters. We entered in last year into a clinical trial agreement with Pfizer to get access to temaciclib and earlier this year initiated that triplet development as well. Sure. Beyond breast, where do you see 2608 applicable? Yeah, there are mutations in PI3K alpha really across all tumor types. You see it in gynecologic malignancies, so cervical cancer, ovarian cancer, colorectal cancer frequently has PI3K mutations. Although in colorectal cancer, the PI3K mutations are frequently seen in conjunction with other oncogene mutations, including KRAS mutations. We anticipate in some of the tumor types outside of breast cancer, there likely would need to be combination therapy that would be used to be able to target both RAS pathway signaling as well as PI3K pathway signaling. That's why breast cancer for us is really the first indication that we're going after heavily because you do have existing proof of concepts here. In PI3K mutated breast tumors, the PI3K mutation happens very early in the genesis of the disease and really is the sole oncogene driver that you see in these patients. Thank you, Don. How is 2608 administered to patients, and what's the current dosing regimen that you've got that you're using in clinical trials? It's an oral drug. We currently are administering RLY-2608 on a twice-a-day schedule continuously, so there's no planned breaks in treatment. In contrast to Capivasertib, which can't be administered continuously, patients take it for four days and then have to have a three-day drug holiday for resolution of toxicity. All of the dosing that we've shown so far has been using fasted dosing, which is very typical as you take a molecule into the clinic for the first time. You instruct patients to not eat before taking the drug. The data we've presented has been at a 600 milligram fasted dose. Last year, we did run a food effect study. There was a small positive effect of taking RLY-2608 with food on the blood concentrations that were achieved. We were able to show that a 400 milligram dose of RLY-2608 taken with food gives us the same blood levels of RLY-2608 as when it's administered 600 milligrams fasted. We took those data to the FDA. The FDA agreed, based on what we showed them, that a 400 milligram dose with food was the appropriate dose to take into our phase 3 trial. That's the dose that we're currently looking at. Wonderful. Last question on 2608. I don't know how much focus you put into this, but looking at street estimates for what the current market opportunity is and flavor probability of success, I don't know if you have any commentary. Yeah, there's maybe about 13 covering analysts at the moment. I would say the variability in those models is quite broad. Of the subset that I think have spent a good amount of time putting some precision to those estimates has bought into it. The range is, again, in a peak sales range, just in the second-line metastatic setting of about $1 billion. Probably success ranges somewhere between 40% and 65%. Wonderful. Thank you. I think that's for the breast cancer. I don't know if you, I don't know if we're going to cover the vascular malformation. Yeah, I'm about to cover the vascular, but yeah, in breast, and we can. That's just the post-CDK4/6 setting. When you get into the frontline, CDK4/6 naive setting, that grows by an order of magnitude. Perfect. Okay. I'm moving on to vascular malformation. I guess how does 2608's mechanism as a mutant-selective PI3K alpha inhibitor specifically target vascular malformations compared to non-selective inhibitors? Yeah. Yeah, so vascular malformations are, it's a non-oncology indication. These are overgrowths, tissue overgrowths, that inform lesions, but they're not metastatic lesions. It's not a solid tumor. In a large proportion of patients with vascular malformations, they have a PI3K mutation that's actually driving the malformation. It's the same PI3K mutations that you see in solid tumors, but this is now in the context of a normal genome. You don't have loss of tumor suppressors or other oncogenes that give you a metastatic tumor. You just have a PI3K mutation that's causing overgrowth of tissue. The way the disease presents is a function of what tissue the mutation arises in during embryogenesis. You have some mutations that could occur early in embryogenesis where you may have a whole limb that's affected where the whole limb overgrows. That is one presentation of a disease subtype called PI3K-related overgrowth spectrum. 100% of those patients have PI3K mutations. There's on the order of about 5,000 of those prevalence in the U.S. Other manifestations include lymphatic malformations and the mutation happens in the lymphatic precursor. About 80% of lymphatic malformations patients, about 80,000 patients, have PI3K-driven lymphatic malformations. Depending on where and when during embryogenesis it happens, you get that mutation. Some patients may have several malformations. Patients may have malformations that are impinging on critical vessels or critical organs and really require systemic therapy to be able to address those diseases. There's venous malformations that have a lower percentage of PI3K mutations and cerebral cavernous malformations where about half of patients have PI3K mutations. PI3K clearly is the driver in these patients. The patients who need systemic therapy need to start it early in life. This is a congenital condition and need to take it chronically, most likely over the course of their lives. You're treating children and you're treating children chronically. Alpelisib, marketed in cancer as Piqray by Novartis, has been approved with an accelerated approval in this PI3K-related overgrowth spectrum or PROS. It's marketed there as Vijoyce, so same API, but marketed under a different brand with different pricing. That has really established proof of concept for how you can actually control the growth of these lesions if you're targeting these patients with a PI3K alpha-targeted therapy. The challenge with a drug such as Alpelisib, it's not selective. It's also hitting wild-type PI3K alpha, so you see a lot of those toxicities that you see in cancer. In addition, you're in kids and you see slowing of growth as you're inhibiting wild-type PI3K alpha. There is some real concern in this population that needs to have chronic therapy that you actually can't or don't want to keep these kids on therapy chronically. This is where we think being able to come in with a mutant-selective inhibitor such as RLY-2608 could be able to leverage what we already know about these lesions and these diseases, that hitting the mutant PI3K alpha driving the tumor can give you control of the lesions while avoiding some of the toxicities that are associated with hitting wild-type PI3K alpha. It's really the same hypothesis that we've already proven out in cancer. We have initiated the trial to prove that out in vascular malformations as well. Wonderful. Thank you, Don. What's the current status of your phase 1/2 trial for RLY-2608 in vascular malformations? Yeah, so we initiated the study late Q1 of this year. One of the benefits of having profiled 2608 so extensively in cancer patients is that it has allowed us to, instead of going to the traditional dose escalation, as part of the dose confirmation in the study, we are able to go directly into randomization of three biologically active doses in parallel. That'll include the recommended phase 3 dose in oncology at 400 milligrams BID fed, and then two doses stepped down from that. We are actively enrolling patients, happy with how the pace of enrollment and excitement amongst physicians is currently going. Not yet in position to guide specifically to when we'll have data yet. Like I said, very happy with how the enrollment is going. Wonderful. Thank you, Pete. Can you sum the population for us in this indication and, you know, maybe help us map out again what the patient opportunity or population opportunity is for you? Yeah, at the highest level, the opportunity is, downplaying out to be at the top. About 170,000 of these patients have PI3KCA mutant-driven disease in the U.S. alone from a prevalence standpoint. As you get into some of the subsets of the disease, about 5,000 to 15,000 of them exist in PROS, so the PI3KCA-related overgrowth spectrum. 100% of those patients obviously have PI3KCA-driven disease. Within lymphatic malformations, there's about 80,000 patients in the U.S. that have lymphatic malformations, and 80% of those have PI3KCA mutant-driven disease. That's about another 60,000. That's the core focus of our initial clinical development. As you can imagine, as a chronic therapy to be used over the life of these patients, even just a few thousand of those patients being on therapy for their lifetime is a multi-billion dollar opportunity for us. The open question, given we have only one molecule approved in just the PROS subtype and only with accelerated approval, not a lot of clinical experience to clinical data to, and therefore commercial experience to reference, the open question here is just how many of those 170,000 patients would seek or be eligible, if you will, for chronic systemic therapy. If it's even just a fraction of that number, it's an extremely large opportunity for us. Thank you. I guess we've kind of covered this on the breast discussion, but just to sign off, how might RLY-2608 safety profile and breast cancer trials inform its potential use in vascular malformations, particularly regarding hyperglycemia and the other side effects? I think as Don Bergstrom covered earlier, these patients are going to be on therapy for the whole of their lives. These are not end-stage metastatic breast cancer patients, so the tolerability profile really does matter much more here. Having a mutant-selective inhibitor that can dial out some of the toxicities that we've talked about—hyperglycemia, diarrhea, rash—is much more important. That's why we're so excited about the use of RLY-2608 in vascular malformation. Wonderful. Thank you. I'll maybe ask a few questions about the platform. I guess how does Relay Therapeutics' Dynamo® platform integrate computational and experimental approaches to address previously intractable, inadequately addressed protein targets? We have been at this really since 2016. We were one of the first companies to enter this space of using computational tools and combining them with experimentation to try and make the process of the discovery of new drugs both more efficient and more effective. I think as we've talked about, each of the many hundreds of steps inside of this process can be broken down and each of those steps can be enabled in some way by computational tools. We break it down into three big groups. How do you identify a modulation hypothesis of where to try and bind a small molecule to a protein? That's one area that we can use computational tools. There we've used long-time scale molecular dynamics simulation to simulate how proteins move and how they can interact with a small molecule to identify novel binding sites, exactly as we did with RLY-2608 by identifying a novel allosteric binding site. The second step is how do you identify novel binders? There we use traditional experimental approaches such as hit finding techniques that everyone has access to. We also have, over the last 10 years, used a range of virtual screening techniques, swimming simulations, and a variety of virtual libraries to find novel binders. The final piece of our process is how do you optimize these binders into essentially drug-like molecules that would eventually become the development candidate. There we've used a range of algorithms, generative AI, to identify novel modifications that we could make with molecules that eventually lead us to the development candidates that we've been able to deliver. It's lots of steps, each one slightly better given the computational tools that we have. String it together, and you get development candidates. Wonderful. Thank you. I just realized I didn't quite finish off on the vascular malformation piece. We talked about the estimates for breast, yeah, around about $1 billion peak sales, but in vascular malformations, maybe the same question. Yeah, I think today it's a bit early to try to estimate that, just given that we, again, there's not well-precedented molecules in the market today. Depending upon the number of patients that ultimately seek chronic systemic therapy over time, it could easily be a market opportunity that is in the single digit billions of dollars. Wonderful. Thank you. Okay, I'm maybe back to the platform. Sorry for jumping around, but what role does the Anton II supercomputer play in Relay's Dynamo® platform, and how might the upcoming exploration of the DesRes agreement AULIS-25 impact their drug discovery capabilities? I think it was very helpful to us over the first, you know, iteration of Relay. The power that Anton II had was unsurpassed at that time. It helped us with a lot of the discoveries that we made in the first years of Relay. Now, obviously, computational power 10 years later has moved on a lot. You can do a lot on GPUs. Obviously, the cloud has moved, so you can split up problems and spread it across the cloud, which is much more commoditized now and allows us to do a lot of the things we used to do on Anton II now, using commodity hardware. Wonderful. Thank you. Final question for you, what, I mean, there's lots to talk about, but what didn't I ask in our discussion that perhaps I should have? I think the question that we get asked a lot is, what does it all mean? At the end of the day, does this whole use of computational technology make the process more efficient? Our answer to that is it's less about efficiency. It's more about output and productivity. In the end, we've been at this now almost a decade. We have put multiple programs into the clinic. We have a phase 3. We have another program that will hopefully file an NDA, that we outlicensed. For us, it's less about efficiency. It's more about output of high-quality molecules. We feel that we've been able to demonstrate that. That's why we're so excited about the potential for RLY-2608. Wonderful. That might be a nice place to park the discussion, but thank you, Sanjiv. Thank you, Don. Thank you, Peter. Thanks, and thank you to Morgan Stanley again for the invitation. Thanks. Wonderful. Thank you, John. Thank you. Thanks.