Schrödinger, Inc. (SDGR) Earnings Call Transcript & Summary

All right. I think we're going to get started here. Thanks, everybody, for joining us. Welcome back to the 46th Annual TD Cowen Healthcare Conference. It's my pleasure today to be joined on stage by the entire Schrodinger team today, minus Karen, who is certainly missed. But maybe we can kind of just go down, Ramy, on my right as the CEO of Schrodinger. And on the far right is Richie Jain, the CFO. And then replacing Karen today -- nobody can replace Karen.

So maybe I want to kind of keep this as interactive as possible here. I will be checking my phone for any questions that kind of come in at [email protected]. But maybe I do want to kind of kick us off with the conversation, Ramy, give you a minute, talk about more substantive update we got last week when it comes to kind of the shifting towards hosted services. ACV reporting relative to revenues. Ultimately, what that means and kind of what all of us in this room should keep in mind?

With regard to the transition to hosted, yes. So I think I'll hand it over to Richie in a second. But the most important thing to understand is -- well, there are a few things. One is, this is a transition that we started a number of years ago. And it's been pretty successful. We already have -- roughly 1/4 of our revenue is already hosted. And the way we deliver the software to the customer, or the experience, let's say, even that the customer has is exactly the same. The price is the same. The way actually even the contracts are done. They pay upfront, and then they have a license for a year. It's just an accounting difference where the revenue, if it's hosted, is recognized ratably over the term of the contract versus if it's on-prem, where it's recognized mostly in the quarter that it closed. So we think this is actually good for customers. It's certainly a better experience for customers. If the software is hosted, we can support them better. We can see what they're doing, monitor their usage, and a lot of our customers are always bumping up against licenses. So that's something now we can see and we can talk to customers and say, "Hey, you're not utilizing this, or you're bumping up against the licenses. You might want to think about purchasing more licenses." So I hope that gives the highlights. But Richie, anything you want to add to that?

Just to add, we look at the business on a cash flow basis, and this change has 0 impact to cash flow. And so because of that, we felt like ACV as an operating metric gives the best sense to investors on how to measure us this year, given that the revenue will have a lot of noise introduced to it because of the accelerated change to hosting. So ACV is a metric we went out with, and it's the closest metric to how we actually run the business.

Okay. And maybe just because you all mentioned this, I think, a little bit on the call, just to kind of clarify, so the kind of revenue and margin impact into the rest of this year and kind of through into next year, how should we kind of just think about that transition too?

Yes. So again, from a cost of goods sold, operating expense point of view, there's no change to the dollars. We do expect revenue to decline this year given most of the business is booked later in the year. And as we transition that over to hosted, you just have less days, less weeks, less months to recognize the revenue for this year. But this will all -- whatever revenue is not recognized this year will be recognized in deferred revenue and will be revenue to be recognized next year. So because of the -- but just mathematically, because we expect revenue to reduce this year, gross margins, our adjusted EBITDA will be impacted just numerically for the same reason.

I think it's also important to point out, if it's okay, I just want to say one thing. This is industry standard. This is a transition that most software companies are undergoing. So this is well understood. The impact, that's temporary, it has on revenue is well understood. A number of companies have gone through this. And this is the way you need to use some kind of metric like ACV to track the growth. So we're not doing anything that...

And just to add on that, the more effective we are in converting this year, the lower the revenue will be, which is obviously a little counterintuitive, but that is the way the math works, and that's actually the way -- what you want. And so given that dynamic, we chose to focus the guidance this year on ACV, because that is, again, truly how we run the business, and it gives you the best long-term view on where the growth is. Given the revenue decline that we expect this year, that is a snapback, and we expect that all to be picked back up in the '27 reporting year. So it's a multiple year transition, but the first year will be the most off the path you'd expect to see.

Okay. Understood. All right. So I know a lot of this kind of comes against the backdrop of this transition and kind of the overall business strategy for the company, right? So maybe let's talk -- start at kind of high level, but then how this kind of feeds into the strategy overall, really kind of just the state of the computational platform today. I guess where are you kind of seeing all the changes and evolutions you've made in recent years? What is kind of drawing new customers to you? And what is kind of keeping customers coming back for more as you undergo this transition now?

Yes, that's a super exciting thing that's happening now. I don't know if people fully appreciate how extraordinary advances have been made in computational chemistry, in the ability to actually run a computation, run a calculation that displaces an experiment that you don't have to run the experiment. That's an amazing thing. So we're doing things now that weren't possible even just 5 years ago. So using the physics engine that we've developed, we can predict so accurately key properties of molecules that you don't have to run the experiment. That's an amazing thing to be able to say. We've been doing this for a long time -- we've been dreaming about this for a long time, and now it's possible. Now the incredible thing about this is that we can run those calculations on a scale that is dramatically higher than anything you can possibly do experimentally. So you have the accuracy of experiment, but you have a scale, let me put some numbers to it. We can, in 1 day, generate as much data as it would take 10 years to generate that data if you did it experimentally. Now what does that mean? That means now you have this extraordinary amount of data way more than you can possibly produce just using experiment that you can use to train AI models. So AI models are now allowing us to scale the physics to even larger space and allowing us to now explore massive amounts of chemical space. What does that mean? That means that we are now, for the first time, able to significantly accelerate the time it takes to get to a development candidate and improve the probability of actually getting there. And then maybe the most exciting thing is you're doing that now with much higher quality molecules. Because you've explored such a huge amount of chemical space with very high precision that the probability of success in the clinic is much higher. Now those aren't just words. You're hearing a lot of these words from a lot of places, and it's hard to figure out the signal from the noise. But we've been applying this technology now for a number of years. And the result is quite a number of programs that have actually gone into real programs. This isn't just saying we've solved drug discovery with AI. We've produced 16 clinical assets, a number of them in late-stage clinical assets, for which we have royalties and milestones. And the newCos that we've cofounded have had really tremendous success, a rate of success that is definitely better than the industry average. So I hope you understand what the technology is doing, how we're able to combine this physics engine with the scale of AI and actually delivering valuable assets. And now one of the most exciting things now that is happening is advances in agentic AI that's allowing us to scale this now in a really dramatic way. One of the limitations in applying the technology is experts to be able to use it. And we're pretty excited about advances in workflows, but also in agentic AI that will allow us to really scale this platform. I think that's what's bringing -- all of that is what's bringing customers.

Yes. And I think -- I mean, this touches actually on a few conclusions we had from -- last night, we had a panel with the R&D investments in healthcare AI. We had the heads of R&D from Novartis and Takeda and Relay Therapeutics as well. And Fiona from Novartis, she gave you guys a great shoutout.

Yes, that's great.

Yes. So I mean I think it gets at some of the disconnect between some of us on the outside who are not using a lot of these tools every day, and we just kind of see pharma pouring more and more money into what the rest of us are kind of collectively deeming AI, but ultimately what that impact means. I think fundamentally, some folks assume, well, if they're spending more internally on AI, that means that there's less dollars that they're willing to spend externally rather than what it seems to be that they're now identifying where the holes in their data are and who they need to kind of turn to help plug those holes and train the models.

Exactly. I think there's been a -- this is well understood by the experts. It might not be so well understood by investors and the general media, but it's very well understood that these pure AI models that are trained solely on experimental data are very limiting, because by definition, drug discovery is about finding new chemical matter, right, novel IP. And that means that predictions of those molecules will not work because they're not in the training set. So you need physics. You need first principles methods to build those training sets. And as I said before, you can do that now, thanks to advances not only in the physics-based methods that we've developed, but huge advances, thanks to NVIDIA in hardware and then, of course, the ability to scale all of this with AI. So...

Pretty exciting. I mean -- okay, so we talked kind of a little bit about the actual transition and reporting for the software business. But all things else considered, how should we think about maybe the next 12, 18 months? Any kind of important inflection points in the actual growth of the software business itself when you have the predictive tox offering is now in beta testing and remains widely...

Yes, it's released.

So as we kind of think about your assumptions for guidance within the context of ACV revenues, but realistically just actual growth of the business over, let's say, the next 18 months?

Yes. So first of all, predictive tox is a really big deal. This is one of the great challenges in drug discovery. It's very common for a drug discovery program to run for 5 years and you do all of this work and spend all this money to identify a development candidate and then you start going into tox studies and you discover that there's a tox problem, and that's the end of it. That's it. You don't know how to fix it. You have no idea where the toxicity. So it's a major source of failure. So this is a big deal. We have launched it. It's come out of beta, because the beta feedback was really incredibly positive. There's a huge amount of excitement, obviously, in solving this Grand Challenge problem in drug discovery. So we expect to see growth from predictive tox this year and into the future. There's a lot more work to be done. There are, in principle, 20,000 proteins that you don't want to bind to, right, the proteins in the human genome. And there are some estimates that it may be a bigger number. We're at roughly 60, 70 off targets that we've enabled in this predictive tox panel. We're adding more as we go. So that's going to be a big area of research and of growth. Now the other thing is what we were talking about before. Not every pharma company -- you met a few of them that are using the technology at scale to generate these training sets for AI, but not every pharma company is doing that yet. They're using it. All of them are using it, but they're not using it at full scale. So we're expecting that to change. We think there have been enough companies that have transitioned to the sort of large-scale use of these methods that it's sort of derisked. And now it's just a matter of more companies sort of adopting the technology at scale. So that's another major source of growth this year in the time frame that you said and beyond, just the actual continued scale-up of the usage. The other area that -- a couple of areas we're excited about, too, is biologics. Our platform has largely been developed and validated in small molecules, but obviously, biologics are very important. We've been putting a lot of work into that, both on the informatics side, but also on the physics side, and there have been a number of advances there that we expect to be able to contribute to growth. And then if I can just touch -- I know this is a health care conference, but I'll just say just very briefly, these physics-based methods can apply to other systems as well, because physics is physics. And so we're pretty excited about the work that we're doing in battery chemistry. Again, I won't spend a lot of time on this, but there are very interesting material science applications in pure material science, workflows like in design of batteries. But in pharma, formulations is a material science problem. And we have new products in that space as well, in particular, crystal structure prediction, which is an incredibly important part of formulation and drug discovery that we also expect to contribute to growth this year. Pat, did you want to add? Is there anything else? Did I cover everything?

Yes, a lot of it. I think one place that we're investing in, too, is beyond just the physics simulation, but we've had the LiveDesign platform that Novartis talked about using to own data. It's really important to be the central platform of drug discovery that allows you to -- anyone who's building any AI models, you want them using it through your platform. Then you become that central hub, which we've successfully done for the strong majority of pharma in the small molecule space, but we have introduced now a large molecule offering. It's especially tempting for people in ADCs and peptides, which are a little hot these days. But because both small molecule and large molecule technologies may be necessary for these, we're uniquely fitted that we understand both of these very well. Most software companies don't. And we're very excited about our LiveDesign for Biologics application. We think that's a great growth opportunity.

Brendan, I'll just add, we're an R&D company. We invest in R&D for our customers who are doing high science R&D. And the investment is to expand our addressable markets. So within life sciences where we've existed predominantly, a lot of the new products we're introducing are immediately adjacent to our customers today, but touch new budgets, touch new capabilities. And as Ramy mentioned, in material science, there are endless end markets there. So we're really excited about the opportunity. We don't typically talk about new products before they're released. Predictive toxicology was an exception to that, just given the amount of industry attention on it and the FDA mandate around new approach methodologies. But some of the other products that we've rolled out on our call last Wednesday are the way we typically release, which is we develop the product and then we launch it to customers.

And I'm glad you bring up the NAMs, right, because it's kind of next natural question that comes out of this a lot. So maybe just help us understand like where this predictive tox offering is actually situated within the FDA initiative, right? And then maybe in that same breadth, when -- I guess I should first ask, have you gotten new customer inbounds or from existing customers specifically tied to some of those initiatives, and when we should realistically think of the impact of that to software growth?

So at the moment, of course, the computational methods that have been developed for predicting toxicity are wanting. They're not very good. And so there's a very heavy reliance on doing it experimentally, which is time-consuming and expensive. What does that mean? It means it gets done pretty late in the process, as I was alluding to earlier. And then it's just -- that's it. Program is dead. You just lost a huge number of years, and that's a real issue. The methods that we've developed, again, as I said really earlier, are highly accurate, really predictive. They predict whether a molecule will bind to one of these off-targets, so-called off targets that are associated with toxicity. And what that means, of course, because it's a lot faster and a lot cheaper to do it computationally and experimentally, it can be moved up really upstream, really early in projects. So in other words, it becomes part of the multiparameter optimization of a molecule. You do it really early on and you make sure that by the time you get to the end, when you get to a development candidate, you've addressed not only affinity and solubility and permeability and so on, but you've also addressed selectivity and therefore, toxicity. So I think there are 2 applications. One is new. It's a completely new market, right? People using this early in discovery. And then, of course, it's still really valuable in the later stages when you're starting to think about what molecule to put in animal studies, and that's where it ties into the FDA. Now the FDA is saying they want to eliminate animal testing. I think every time somebody hears that word, they think, come on, that's crazy. But it's okay. It's okay to think crazy, because the future isn't so far away. Right now, it isn't going to eliminate animal testing. But it's clear, and it has been reducing it, because, of course, if you have a molecule that's lighting up in this computational assay and saying it's going to be toxic, why would you put it into an animal? So it will certainly result in reducing animal testing. Maybe in the future, it will eliminate it. That seems really far-fetched that you would actually use humans to test the toxicity, but it's okay. You get the idea. It's going to significantly reduce it. So I think those are the 2 applications.

And one other thing I'd elucidate is that it's not just for animals, we can test both the animal and the human protein. So you might go through the entire development process just looking at the human protein and then it fails in animals and you're like what just happened. But since you can test both of those, you'll be able to uniquely identify ahead of time. If you see something in animals that's different than what you're seeing in humans, you might be able to know ahead of time and expect that or vice versa. It looks fine -- in the worst-case scenario, it looks fine in animals and it has a problem in humans. Knowing about that earlier, obviously, is incredibly valuable, because clinical trials are even more expensive than the animal studies. So that's extra knowledge that just largely doesn't exist right now. There are some experiments to try to get at it, but that level of information and be able to push that early should dramatically increase the success.

Got it. Okay. So when we kind of think about now continued evolution of the platform, and I promise you all continue to come back to this question over the months and quarters ahead. But now that you're all kind of transitioning really to a fully-fledged software entity with a few notable exceptions around the edges there, how do you kind of think about continued evolution of this, right? Obviously, we have predictive tox, but you mentioned biologics, you mentioned a few other modalities. Is it kind of order of operations to expand within what you've got to other modalities and then maybe to other parts of the drug development spectrum. Like where does that kind of strategy fall?

Yes. As Richie said, we're an R&D company. We have a significant investment in the platform. One of the areas that we're super excited about right now, and it's actually the -- what I'm about to describe is a technology that's actually enabling the predictive tox initiative. And that's, in general, just protein structure prediction. So there have been a lot of advances in computational methods and experimental methods for determining the structures of proteins. You heard about it, AlphaFold. I mean there's Nobel Prize, right? But what's not so well known is that the output of those initiatives, the experimental and the computational, is pretty low-resolution structures. They're not actually that useful out of the box. We are developing methods for refining those structures to high resolution, which is actually what you need to make use of them. You got to get the details right. If you hear me say the word physics, right, physics-based method, you can imagine the input to a physics-based method is getting the positions of the atoms in the right place as a starting point. And so that's pretty important. So we're putting a huge effort into determining the structures of proteins and the molecules that they're bound to, to high resolution. Now what does that do? At the moment, we really only know the structures of proteins, human proteins to high resolution of maybe 10%, 15% of the human proteome. Obviously, the ability to scale that up to 100% allows us to work on targets that we otherwise can't work on, the so-called hard-to-drug targets, targets that are implicated from the point of view of biology in important diseases, but we just don't know how to target them. Is it with a small molecule? Is it a peptide? Is it a degrader? If it's a small molecule, is it a macrocycle? Is it small? Is it big? Is it -- right? All that sort of thing. So enabling us to actually explore all of biology through knowledge of the structure is huge, to really open up -- and again, that's the technology that's being used to enable us to be able to predict binding to off targets. But obviously, identifying or being able to design molecules for targets of interest from the point of view of solving diseases is obviously a really big -- is important and a big area of research.

Yes. And I just want to add on, too. We have the ability to see our most popular workflows that are used in our software. And by far, our most popular workflow is one that takes PDB structures and cleans them up, because your average PDB structure is so far away from being usable in drug discovery. And I think this is really important because most of these AI models are trained to try to reproduce the exact PDB structure, which our customers are telling us, through that utilization, are not good enough. So the best case scenario is they're reproducing at the same quality that is not good enough for use, and that's why we really invest in that.

And PDB structures, that's the structures in the public domain.

Yes. And I guess kind of tied to this now, before we get into the therapeutics pipeline just in the last couple of minutes, I did want to ask the partnership strategy. I know you all kind of announced a new partnership with TuneLab over at Lilly. And I think it's kind of more focused on this idea of kind of federated learning, right? But maybe help us understand how that approach and that specific partnership relative to like the Novartis, right, that we talked about before, how does this all now fit into Schrodinger's overall partnership strategy and where that kind of fits into the growth story for the software business overall?

Yes. We're super excited about TuneLab. TuneLab covers a huge gap that biotechs have. So for as long as companies have had LiveDesign, which is approaching 15 years now, which is kind of crazy, every big pharma has put in these machine learning trained tox models built on all of their data. They've gotten a little better at it, but really, it's just kind of around the margins. And one limiting thing is how much data they have. But what we see happening is when people leave pharma and they've gotten used to doing drug discovery with these ML tox models that they've built, they go to found a biotech and there's nothing. There's no public because this is all built on their internal things. So what Lilly has done, it's awesome here, is they've figured out how to give every biotech out there access to those types of models. And obviously, Lilly is not just doing it for fun. The biotechs then put their data back in and Lilly's own models get better. But it's super exciting for us because selling LiveDesign this entire time, the first question is, do you have any suggestion for how I get this model like I had back at Big Pharma X? And our answer has historically been no. Now I do want to address the elephant in the room, because we get a lot of questions, isn't this directly competitive with predictive tox. It's not. So the accuracy of these type of models is totally different. They're very useful. They're often on endpoints. They're much higher level than what we simulate, but the correlation between the endpoints is much lower. It is still useful, obviously. But when they're using our engines, they typically expect an accuracy that they can make confident decisions in. And these are more kind of like red light, green light hinting accuracy. Still very useful. I don't mean to denigrate it at all, but it's just a different tier.

I just want to add to your kind of long-term growth strategies and how we partner. What we see as a long-term driver is enabling our users to become power users. We have throughput-based licensing and pricing. So the more any individual user uses, we are able to capture that value. But as the workflows become more efficient, we can enable our users to be able to run more and also expand the amount of users who can run our technology. So Pat is actually working on a lot of the integration with LLMs and other agentic AI processes that will be able to expand our user base over the years.

Okay. Great. So I think now in the last few minutes here, I did want to touch a little bit on kind of the status of the therapeutics pipeline, right, both internally and partner here. So maybe just quickly give us a sense of when we could get updates from 3515 and 1505, that's the Wee1/Myt1 and then MALT1 inhibitor, respectively. And ultimately, kind of what the status is of both assets, whether that's external licensing and partnership discussions, where that all stands now?

Sure. I'll address that. So our intention is to finish the dose escalation studies on both 1505 and 3515. We presented data on 1505 last year. 3515, we should be presenting data in the second quarter of this year. But importantly, we've announced that we see the best way to advance these assets are with partners in the mid- and late-stage development. So that's where the focus is. We'll give updates as we have them. More broadly, as we think about therapeutics, we continue to be really excited about the collaborations portfolio, working hand-in-hand with our pharma partners and not only generating IP and delivering development candidates, but enabling broader adoption of our software within those customers and also generating downstream milestones and royalties that we're accruing at this point. And we're excited about the targets and the indications and the royalty rates that we have there. We gave some additional disclosure in our results last week to kind of give a sense for what that opportunity is. But 5 of these programs are in $5 billion-plus markets where we have royalties ranging from high single-digit to low double-digit ranges.

Okay. So it's fair to assume now kind of in perpetuity moving forward, at least for the foreseeable future, that any new therapeutics, new drugs that could come out of the Schrodinger platform would largely be relegated to existing partnerships that you have with pharma biotech externally, right?

That's right.

All right. So I guess just in the last minute here now, I want to kind of pull everything home a little bit. So we've talked about the transition with ACV. We talked about therapeutics pipeline. We talked about the partnership strategy and evolution of the platform. So with all of that said now, as you kind of look ahead, not just into the latter half of this year, but really into kind of this next era for Schrodinger, where is kind of the biggest disconnect when you talk to folks who are trying to understand where you all fit in, what the real growth drivers here are and ultimately, where they are kind of trying to value the platform?

Yes. I think the biggest disconnect, and you can't blame people for this, is there's a lot of noise out there. I mentioned it earlier, right? There are an uncountable number of companies labeling themselves as AI companies that have completely changed drug discovery, have solved the problem. And they're publishing blogs and publishing white papers and doing comparisons and getting a lot of attention actually. And it must be really overwhelming for -- I mean, how are use supposed to do unless you're an expert in all things, physics, chemistry, biology and computer science, to figure out the signal from the noise. And what I would invite people to do is just look at the track record. That's important, right? What have these companies actually produced? What have we produced? And I think if you do that, there's a very clear distinction between companies that are just saying that they're doing things and running retrospective analysis, right, running calculations on things that are already in the literature and saying, look, it matches up. As everybody knows who's developing machine learning and AI, that's really easy to do, because those things are in the training set. So that doesn't count. And if you haven't produced development candidates and clinical assets and you don't have 100% customer retention from customers, I'm not sure you should be out there doing all of that. So sorry, I know that sounds a little bit critical, but you can imagine the frustration from a company that has been doing that for a long time, has had a track record of delivering over and over again high-quality clinical assets that are progressing through the clinic, for which there are very meaningful milestones and royalties, by the way, on quite a number of them. And the success of the companies that we've co-founded is striking. I mean it's noticeable. That's not normal for that sort of success rate and then the size of these exits and so on is really meaningful. So I hope I keep saying it and having an opportunity in venues like this to highlight it. I would encourage people to go look at it, look at the pipeline, look at the success. I hope you will see a difference. Don't take our word for it that we're just saying we have a physics engine that's accurate. I mean, look at the results. I think that's really important. And I think the other thing is, don't be -- it's very dangerous when there's a new technology, it's so easy for it to get overhyped. You have this tendency to sort of extrapolate in the future and say, well, I don't understand this stuff, but boy, it sure looks like it's going to be able to do something that is magical and I don't really understand. Usually, that isn't the case, right? It's a technology like everything else. It has a domain of applicability. It works in certain circumstances. It doesn't work in others. Treat it like a technology, not like magic and something that you should be really afraid of. I don't know if I'm saying anything useful, but it's on our mind. I hope it's useful.

Yes, looking at the actual use case for a lot of the tech resistance. All right. Well thank you guys for joining. It's always a pleasure to see you. Thank you, everybody for listening. We got more to come.

Thank you very much.