All right, well, welcome once again to the 44th Annual TD Cowen Health Care Conference. I'm Yaron Werber from the biotech team, and it's a great pleasure to moderate the next panel with Relay Therapeutics. All the way to my left, really needs no introduction, is Sanjiv Patel, President and CEO. To his right is Don Bergstrom, who is President of R&D, and to his right is Peter Rahmer, Chief Corporate Development Officer. So, gentlemen, thanks for joining us. You know, AI has been a big theme for us at TD Cowen for a while. Can, can we talk? We all know about the Dynamo platform. Maybe can you talk about how you've been modulating and fine-tuning the platform over time, and how much learning, new learning, and new capabilities are you developing?
Thanks, Yaron. And thanks to Cowen for the invitation. Congratulations to you, Yaron, for climbing Kilimanjaro and doing it in a Relay hoodie, so very excited about that. Back to your.
Thanks. It kept me warm.
Back to your question. As you know, it's been in the, in the news and the popular media now for well over a year now around AI, and obviously, is topical. I think we've been at this now for seven or eight years, trying to meld computational techniques, with experimental approaches, and then combining that with, automation to try and make the discovery of medicines, more efficient and effective. And so to answer your question, our platform is really focused on, making medicines as opposed to, you know, there's a lot of hype around, discovering new targets and elucidating novel biology. That's not where we focused on our seven or eight years. It's really been on, making, small molecules, bind to proteins of interest and make them do exactly what we hope that we, we want to do. And so we focus on validated targets.
So in terms of, you know, what improvements we've made over those years, I think the most kind of stark thing is obviously we've put four molecules now in the clinic. At least two of those now have shown validating clinical data. And the third of those, our SHP2 molecule, is with our partner, Roche/Genentech. And obviously we are very hopeful that that too will show validating clinical data. I think in terms of where the improvements are going, I think it's really just this combination of the three elements. And over time, we've got better at combining and understanding which computational approaches fit which task. And so as you know, the discovery of a medicine consists of hundreds of different tasks. And so it's not just one kind of magic button or one tool fits all. We have to look for the right tool.
Then it's combining it and creating a culture where computational scientists can work hand in hand with experimental scientists. They do that to understand how to modulate a protein, find a starting point, and then how to optimize it. There are tools that we're excited about, like machine learning, dealing with encoded library screens, understanding how to use predictive algorithms to predict DMPK and drug-like properties of molecules. But I think the thing that we have kind of invested in much more over the last few years is automation of chemistry. You can come up with lots of predictions, and then obviously it takes, you know, several months to do the actual cycle of synthesizing and then testing. As we automate more, we can speed up those cycles. So I think ultimately what, what have we been able to do is put molecules into the clinic.
I think that's probably the biggest change over the seven to eight years of our approach.
What about the virtual access to your point on the machine learning automation versus reducing that to actual wet lab experimentation?
So in terms of just the translation from the computational predictions to the wet lab, I think we've been much more adept now at testing a much greater range of hypotheses virtually, and therefore then bringing only a small number of those hypotheses into the wet lab to synthesize, thus reducing costs and increasing kind of the ability for us to actually then reduce cycle times. So I think over time we've got much better and better at this. It also helps us take a kind of a greater exploration of the patent space as well. And as you know, one of the big questions is what are the barriers to entry in any of these things?
I think one of the kind of benefits of using this approach is you can explore a lot more chemical space much more rapidly, and then you can build yourself a much broader patent estate around the molecules that you're really excited about.
And then as I think about Dynamo, one of the key aspects of Dynamo is the protein-protein simulation. How much fine-tuning have you done to that capability?
I think over time, I think we're probably the most experienced in the world of how to use protein dynamics and really understanding the motion of proteins to discover novel hypotheses. And so obviously we've had a partnership over these last eight years with D E Shaw, using their supercomputer. We've been able to obviously develop our own in-house techniques to do long-time scale molecular dynamics simulation. And as you know, given the prevalence now and access to cloud computing, we've been able to then transfer a lot of that molecular dynamics simulation capability in-house. And obviously you can do shorter and shorter pieces of molecular dynamics simulation on the cloud and stitch them together to create these long-time scale molecular dynamics simulations.
And so over the eight years, we've been able to go from being fully reliant on a capability that was really pioneered by DE Shaw to doing the majority of these things in-house.
Okay. And how do you in terms of how do you fine-tune that capability? 'Cause, right, a lot of it this is where the wet lab experimentation, the real-world experimentation, needs to match up to the virtual capability.
Yeah. I mean, to start with, like, there's nothing virtual about the start of this process, which is synthesizing full-length proteins. And this creates, you know, obviously experimental data sets that are driven by techniques such as imaging using room temperature X-ray crystallography, or cryo-EM. And so the basis of these simulations is reality. Then we create these virtual simulations using, the techniques that we've been able to pioneer because we have access to the Anton supercomputers, so we can compare our own, internal stitched-together long-time scale molecular dynamic simulations with the gold standard. And then obviously we make computational predictions, which we then test in the wet lab and make sure that our predictions are based in reality and then refine. So we've had eight years of refining that over across multiple targets.
Okay. Great. Let's move to , your selective PI3Kɑ program. And we've seen some, you know, obviously early data so far. The latest one was at AACR last year. It showed that the, the program was fairly selective. There was no grade three, four hyperglycemia, no AE-related discontinuation. The combo on the efficacy side, the data wasn't quite what we were hoping for. And at the time you were testing the 600 mg and 800 mg BID, and you're now continuing to do a longer follow-up. In the meantime, you've done obviously the fulvestrant combo, and now you've been adding a ribociclib, you know, triple combo. Is 600 mg or 800 mg BID still the dose, or is there a reason to go higher?
So if we take a step back, 2608 is a PI3Kɑ mutant selective molecule. And as we saw last year with Roche's disclosure around their own program in inavolisib, which is a non-selective PI3Kɑ inhibitor, the mechanism is clearly valid. The Roche data showed significant benefit by adding a PI3K inhibitor into the first-line treatment regimen that they had in hormone receptor- positive, HER2- breast cancer patients. So the early data that we showed last year, and we showed two cuts of it last year, really started to validate the mechanism, which the Dynamo platform had helped us predict how to make a selective PI3Kɑ inhibitor, which we believe that no one else had been able to do. And it showed clear selectivity. And that selectivity was manifested in low rates of rash, diarrhea, and then most importantly, hyperglycemia.
The efficacy data, as you said, was very early. And as you know, the approvable endpoint in this field is gonna be progression-free survival. So I think our hope over time was to make sure that that first data disclosure was all focused around safety and efficacy. And over time, we would show, sorry, safety and tolerability. And over time, we would show the efficacy. We showed a later data cut in the year that for the first time showed clinical benefit rate of a much higher rate than we've seen in the non-selective inhibitors, although still early. And so I think over time we'll continue to create data that will give us and you comfort that that this molecule will show a step change in efficacy too.
And in terms of the data updates this year, is it mostly gonna be the fulvestrant combo at the 600 mg or 800 mg doses with a long-term follow-up, or, you know, or, or can you confirm whether you are indeed actually testing even a higher dose?
So maybe, Don, you take the dose question, which I didn't answer previously, and then Pete, you take the data question.
Right. So in August 2023, we disclosed we were opening the 600 mg BID expansion cohort while the dose escalation was continuing. And then in January 2024, just a couple of months ago, we indicated that we've now completed dose escalation. We are expanding the 600 mg expansion cohort. So it originally was 20 patients. We filled those slots for 20 patients. We added an additional 20. And at the same time, we have opened a 400 mg cohort primarily for Project Optimus reasons. But we really continue to be enthusiastic about what we're seeing at the 600 mg dose.
I think when we look across the totality of the data, so if we look at the PK data, pharmacodynamic data, early efficacy data, and then safety and tolerability data, the 600 mg dose does look to be an optimal dose and is the one that we are taking forward most heavily now in our expansion cohorts.
For Optimus, that's with fulvestrant?
It is.
All of those, the 600 and the 400?
Yes. Both of those are with standard dose fulvestrant.
The triple therapy, that's still gonna be at the 600?
Yeah. So the, you know, there will be some abbreviated dose escalation work there since we don't have an established dose yet with triplets to maintain some safety margins. But, you know, I think the goal would be to get to a similar dose in triplets.
Maybe, Pete, you can just take the data question.
Yeah. So I think as you alluded to, you're on the main key focus data disclosures this year for us will be continued follow-up of the 600 mg twice daily plus fulvestrant doublet cohort. As Sanjiv alluded to, we've answered the safety and tolerability question fairly robustly with the dose escalation and early expansion data. We've already shown a little bit on the safety side. And so here the main question that we want to—we won't completely answer it, but we'll definitely start to answer in the data disclosure the second half of this year—is what the longer-term efficacy look like in the doublet cohort?
So we think we should have 40 or so patients with at least six months or more follow-up, and with at the 600 mg doublet dose by the time we get to the second half of this year. And there certainly with patients that once they've reached the six-month point or at least had the opportunity to, you can define CBR, clinical benefit rate, which is just stable disease or better at the 24-week time point. And also we might be able to do some landmark analyses on PFS. What does six-month PFS look like? Potentially nine-month depending on the number of patients and follow-up data there.
Yeah.
So being able to show that, and we still have to continue to maintain that safety too and continue to follow safety. And so I think that's gonna be the bulk of the focus on the disclosure in the second half of the year.
Pete, you can just cover the hurdle.
Sure. So what we know today is that if we were to move forward with a randomized study in this setting, this HR+, HER2-, PI3Kɑ mutated metastatic setting at the second line—so after they've been through a first-line CDK4/6 containing regimen—we believe that the regulatory standard of care would likely be a controlled study against either capivasertib, the AKT inhibitor from AstraZeneca, or alpelisib, the non-selective PI3Kɑ inhibitor from Novartis in combination with fulvestrant. In either of those instances, we think that control arm would probably generate something in the order of magnitude of 5.5-6.5 months of median PFS.
Right. So the thought would be to go head-to-head with a primary endpoint of PFS and hope to go to 8.5-9 months?
Again, we have not taken these data to the FDA and had that level of discussion. But based off of the data and the study that we have ongoing, that would be the first study we would likely be able to run, is that second-line study. And it's a study that would be manageable for a company of our stage and size. And we would be eager to run that ourselves if that's how we ended up progressing.
Do you need to also beat that comp on response rates, or PFS is the primary and secondary can be response rate, and that's not as important?
It is absolutely not as important to differentiate on response rate. For a whole litany of reasons on the biology of this disease, but also from a regulatory standpoint, PFS is absolutely the definitive standard of care regulatory hurdle.
Usually response rates and PFS are somewhat interlinked, right? So can you it. I'm just reading between the lines. It might not. There might not be the same relationship in this case. What do you know about what explains that biologically?
Let me take a note.
Yeah. So, you know, I think first of all, the disease context here matters. In the field of the population we're enrolling, there are a sizable minority of the patient population that can be addressed with this mechanism. We don't actually have disease that's measurable per RECIST criteria. These patients primarily have bone disease. And while it can be imaged using the RECIST criteria for addressing partial response, for instance, these are patients who are not measurable for the purposes of calculating a response rate. And so already you know that there's a significant amount of patients who have need for whom RECIST is not the appropriate measure for measuring response. Then on top of that, if you look historically at trials in this in this space, especially in the post-CDK4/6 setting, you see a disconnect between the reported response rates and the reported PFS.
Just within the PI3K pathway targeting class, if you look at the data that were reported for capivasertib, the response rate that was reported, I believe was 29% in their phase III trial with a 5.5-month PFS in patients post-CDK4/6 inhibitor therapy. If you look at alpelisib or inavolisib, in a similar patient population, the response rates for each were 19%, so lower, but the reported PFSs were seven months or eight months, depending on the trial that was run. So you don't see this tight linkage between response rate and PFS.
In fact, in many cases, you actually see that, identifying dosing regimens or doses that could potentially push response rate frequently come with a detriment in PFS, which could be due to the fact that, you know, the doses that maybe eke out a little bit more response for a drug like alpelisib that comes with significant toxicity liabilities that limit the ability to maintain patients with the intended dose intensity.
Okay. When you're thinking about in terms of the IC90s or IC80s and IC90s, something Wall Street talked about for a long time, I think with the 800, you're sort of at the IC80 level, right? Can you correct me if I'm wrong?
Actually starting at 400.
Starting at 400.
Starting at 400, we're getting that coverage. And then we get progressively more coverage at 600 or 800. But 400 is the lowest dose that gives us continuous 80% coverage.
With the 800, what level of inhibition are you getting there?
It's between 80 and 90.
It doesn't seem like, just reading between the lines, going higher makes a lot of sense from the way you're thinking about it.
You know, I think as we characterized our PI3K portfolio of molecules preclinically, we've been able to largely validate the data that were first published by scientists at Novartis in 2014 that identified this time above 80% coverage as being key. So as we entered the clinic with 2608, that was our goal. That was the threshold we wanted to hit. We've hit it with 400. You know, I think while the data are still early, we have seen evidence of clinical activity at 400. 600, I think, is getting us to the point where you comfortably are exceeding 80% coverage in the vast majority of the patients who you treat. I think consistent with that, 600 mg really looks like the optimal dose moving forward.
Okay. You've also decided now to deprioritize 5836, right? That was in the clinic for a fairly short period of time.
Yep.
What led to that decision?
I mean, back when we put it in the clinic, it came in, you know, close to 18-24 months after 2608. And it was designed purely as a hedge against kind of seeing anything idiosyncratic with 2608. Obviously, it's such a significant market, and a company of our size and scale, we didn't want to take any risks. And so we produced a chemically distinct, second molecule. As we kind of learned from the clinic experience around 2608, we didn't see anything idiosyncratic. And so for us, 5836 had to be a kind of step change, better, and to give up that kind of two-year clinical lead that we have. And so again, being capital constrained and focused enough clinical execution, we made the decision, as rapidly as we could, to focus on clinical development on 2608.
You know, it's exactly the molecule and profile that we'd hoped it would be.
Yeah. There are other mutant-specific PI3K ɑs in development. They're not pan-mutant-specific. You know, Loxo obviously and Scorpion. Any thoughts on those molecules? On the concept?
I'll make the obvious statement that neither of those molecules have shown clinical data. So Scorpion, because they've just recently entered the clinic. Loxo with 783 came into the clinic shortly after we did, but to our knowledge, have not shown any clinical data yet and have not guided to showing any clinical data. In the case, there are two different profiles. Loxo 783 is an H1047R-specific inhibitor, so only targets one of the key hotspot mutations. So, and we estimate that's probably, you know, 40%-ish of the mutated patient population that harbor that H1047R mutation. So clearly limited to just a portion of the patient population. Otherwise, you know, again, we haven't seen any clinical data to date. So tough to comment too far.
And then Scorpion, their goal is to have a pan-mutant selective inhibitor, as ours, but have really only focused on H1047 development in their initial clinical exploration. And it haven't shown a lot of supportive data about having activity against helical domain mutations. So that 60%, 50%-60% of patients. And obviously two years behind our molecule.
You also had an H1047R preclinically. Is that still in development, or what's the latest on that?
Yeah. I mean, we still have it in preclinical development. And as you can see, you know, we have a broad portfolio of PI3Kɑ mutant selective assets. And the reason is it's such a big clinical opportunity, which is now clearly being validated, as up to the kind of data we've seen from Roche last year. And so we wanted to have as broad a range of opportunities as we could to be successful. 2608 is showing exactly what we want to do, which is pan-mutant activity. And so really the question is, do we need a 1047 specific molecule? We'll have it if we need it. And you could imagine thinking about using it either in combination or in series. But I think we'll just let the data play out, knowing that if we need it, we'll have it.
Okay.
We'll likely continue research for a very long period of time against this target. So, I mean, I wouldn't. I know folks will, but caution not to read too much into that. We're gonna. We'll stop when we think we've made the definitive molecule to be long-term best-in-class in this area.
Yeah. Can you also had a CDK2 in development, and you had an alpha degrader. And I think both of those have been paused for now. Can you give us a sense, you know, what led to that decision?
Yeah. I mean, we were excited about both profiles. Obviously, the clinical data that we've seen over the last year has even just increased that excitement. I think that what we see in a capital-constrained environment with kind of executional focus that we have was that we wanted to push forward 2608 and 5836 as rapidly as we could. Obviously, we've made the decision that, you know, which one to take forward now with 2608. So that was what was really driving the pausing of those assets. But we remain very excited about each one of those. So each one stands ready to be kind of progressed at the right time. So I think we just look forward to seeing the data play out.
Okay. So you have a fairly broad program. I'll come back to lirafugratinib later on. What I think historically you talked about oncology. And you talked about non-oncology targets with your platform, potential mutant-based, being potentially metabolic or orphan. How are you still thinking about your focus? Is it still mostly oncology or not necessarily?
I think the oncology as we've talked about in the past, you know, our programs consist of, oncology and then, genetic disease. And they have the same characteristics of genetically defined patient population, validated targets, and a rapid, pathway to proof of concept in the clinic. And that's something that, that all of our targets, to date have really been focused on. And so I think when we have committed to is later this year is sharing a new, novel, innovative, molecule that will go into this clear profile. It can come from either side, of our portfolio, oncology or genetic disease.
At that point, that's gonna be a preclinical disclosure or a disclosure of an IND?
No. Most likely, historically, we've disclosed our program somewhere between in and around DC or later.
It's about a year, let's say, pre-IND?
Give or take.
Okay. Any questions from the audience? Okay. Maybe, lirafugratinib . In terms of the next stage in development, the data looked very, very good in CCA. I think you've decided not to continue while you continue to develop that program, not to prioritize that toward commercialization now. The data in the non-CCA fusions was about a 35% response rate. In the FGFR2 amplifications, was 24%, but then specifically 40% in the breast cancer sub cohort and 13% ORR in the mutant population. So we're beginning to realize that, you know, not all mutations are created equal, you know, which was always a hint. But now we're beginning to see that in the data. So what's the next stage? Is it really a pan-agnostic development? And then where does the line of therapy come into play too?
Can I take a peek?
Yeah. So we paused as you alluded to. We paused the CCA commercialization efforts at its inclusive of continued regulatory efforts in the near term against cholangiocarcinoma. Largely given the ability to let the tumor-agnostic development catch up a little bit. I think the IRA kind of put these types of contemplations to a lot of companies. And so it's just a conscious decision we made late last year. And so I think what we've guided to this year is we want to continue to follow the tumor-agnostic data and assess the strength of that data. As you alluded to, it has variability across the different alteration types and potentially even within tumors.
So we wanna make sure we generate a bit more data with more durability and follow-up before we make a decision as to how to utilize resources against that program and the best way to do that. It will likely come with an interaction with the agency at that or around that point in time also. But it is not a, you know, it's been, there's been commercial challenges to these tumor-agnostic labels over the past handful of years. There's definitely been a bifurcation between the ability to get to a label and the ability to commercially execute against that label. There's no such thing, generally speaking. You know, you don't find many treating oncologists that are tumor-agnostic. They tend to specialize in one or the other.
And so you, you're having to find yourself needing to, even though you could get to that label, you still have to generate the data to support broad use across a lot of different tumor types. And I think we're seeing that, it's kind of coming through in the guidance that the FDA has given and how they even want these tumor-agnostic filings orchestrated. So they want good diversity of tumor types. They want consistency of activity across those tumor types. I think it's leading some companies to think twice about their strategy there based off of that.
Okay. And what's gonna be the next disclosure just for that program? Is it gonna be a data disclosure, it sounds like, or?
Yes. We've committed to a data disclosure in the second half of this year.
Across all cohorts?
I think the focus will be outside of cholangiocarcinoma. I think we'll monitor the data that we have in hand before we get too specific about exactly what dataset would be.
Okay. In terms of just new programs, I mean, so we're gonna get one more disclosure this year. Is there any thoughts about having a cadence of one per year or, like, I don't know how much you can handle?
We've always been resistant to do this contrived two INDs a year, one new program a year. That implies you can predict the cadence of the science and the experiments. And that, it seems, we can't do it. Maybe others can do that. So we tend to just let the science play out. And when we have something tangible and robust to be able to disclose, that's how we kind of, that's what dictates that cadence for us. The platform has been exceedingly productive. So in the, you know, first seven, eight years, we've, it's produced, you know, going on double-digit programs, both preclinically and clinically in total. So it is really productive. And we're very happy with the pace of its evolution and its productivity and efficiency.
But I think you'll see us kind of fall short of trying to put a stake in the sand as to number of INDs or programs per year.
Yeah. Just remind us your cash position and how long does that last for?
Yeah. $750 million, as of the end of the year. And continue to guide to that being sufficient to get us, at least into the second half of 2026.
Okay.
A number of catalysts and data points that we can get through, you know, over the 2+ years of time we have with that runway.
Yeah. And so to go back, maybe final question on 2608. After we get the updated PFS data later on this year from the doublet, what needs to happen before you can move on to a pivotal?
One second. Yeah. So, you know, I think it would be taking these data and having interactions with health authorities, to get their agreement on the design of a pivotal trial, which would include what's the comparator, confirming, you know, relevant endpoints again, which all precedents would lead to PFS. And importantly, confirming dose. You know, we've seen in this era of Project Optimus, FDA is very focused on picking really the dose that has the optimal risk-benefit profile, not just an MTD. We think 600 mg is shaping up to have that profile. We've opened the 400 mg, to be, you know, a comparator to that, if you will, to use to justify the 600 mg dose. But obviously, that will be a point of discussion.
Yeah. Well, terrific. I think we're out of time. Team, thanks so much for joining us. Good to see you. We appreciate it.
You too.