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

Great. And we're going to kick off our next session. Thank you for joining us, everyone. My name is Mike Ryskin, Bank of America life science tools and diagnostics analyst. Joining me from the team is the senior analyst, Derik De Bruin. And for our next session, we have Schrödinger with us. Joining us is Joel Lebowitz, CFO; Karen Akinsanya, Chief Biomedical scientists. And we also have Dr. Robert Abel, Chief Computational scientists. And to kick it off, I think we're going to have a brief company presentation just to go over some background and to sort of set the table a little bit. And then we'll go into the fireside chat Q&A format that we've done before. So with that, Robert, take it away.

Thank you so much, Mike. Delighted to share about Schrödinger. So on Slide 2, you'll see typical cautionary notes. On Slide 3, you'll see an introduction to Schrödinger as a company. We are committed to transforming the way therapeutics and materials are discovered. And we achieved those goals by way of our industry-leading computational platform. And that platform enables the discovery of high-quality molecules for drug discovery and materials design much faster at a lower cost and higher likelihood of success than traditional approaches to discovering such molecules. We use this technology to pursue 2 main lines of business. The first is our software business where we license our technology platform to drug discovery companies, pharmaceutical companies, biotech companies as well as materials companies broadly worldwide user base. We also use our computational methods to pursue collaborative and internal drug discovery efforts, which includes around 25 collaborative programs as well as our internal drug discovery pipeline, which Karen Akinsanya, will speak to in greater detail in just a moment. The way this computational technology platform provides value is highlighted on Slide 4, where we have to note at the outset of this presentation, designing drugs and designing materials is extremely challenging. It takes a long time. It's highly capital intensive and one sees a typical industrial setting, very high failure rates. And a big part of the challenges highlighted on Slide 4, is successful discovery requires the identification of molecules that balance a large number of anti-correlated properties in a drug discovery context, this would include multiparameter optimization of things such as potency, selectivity, solubility, bioavailability, clearance, permeability, various other properties that need to be in very tight tolerances in order to have a molecule one could advance into the clinic with confidence. The natural consequence of this challenge is highlighted on Slide 5, where one can see a potent molecule might arise very early in a drug discovery effort. But when one goes in and attempts to improve other properties, for example, selectivity, the anti-correlated nature of these properties means you might compromise potency. And then you have to go back in and try to recover potency as depicted in the third molecule, but then one needs to start in on the other properties, which may lead to backslides of earlier optimized properties. This whack-a-mole problem attempting to simultaneously optimize all of these properties is a major contributor to 2/3 of industrial drug discovery projects ultimately never succeeding in identifying a molecule they can advance into clinical studies. The way we believe this should work is detailed on Slide 6. In the future, if we could calculate all the properties necessary for a molecule to be advanced into the clinic with confidence using computational methods. And further, we're able to apply those computational methods to the whole of chemical space, all synthesizable molecules, then one could immediately find at the outset of the drug discovery project the single best molecule, which would be most ideal to synthesize and advance into in vivo studies and ultimately, clinic. On Slide 6 -- excuse me, Slide 7, we highlight that there's really been 2 major approaches historically to developing such methods that could identify that perfect molecule that could be advanced in the clinic. This is a problem that has been pursued for decades in both academic and industrial circles, and there's really 2 major approaches. The first is knowledge-based methods. These are machine learning methods. They're often, in the last several years, referred to as AI methods. And they attempt to infer from earlier known information, what new molecules might be best to synthesize. And there's been a lot of excitement recently about these methods because of their success doing things such as playing chess, playing golf, doing photo classification. The other approach to modeling these problems has been to develop rigorous physics-based approaches. We're actually calculating in atomistic molecular detail, the property of interest. But in order to apply those methods successfully, one needs to develop a detailed understanding of the underlying physics of the particular molecular property you're intending to compute. On Slide 8, we highlight how we have been able to combine such physics-based methods for property prediction with machine learning approaches in a way that recovers the very high accuracy provided by such physics-based methods while also inheriting the speed, throughput and addressable scale provided by machine learning methods. Through this integrated approach, we're able to evaluate billions of molecules in the chemical space, perform multiparameter optimization to allow for much more rapid identification of high-quality development candidate molecules which then can be advanced into IND-enabling studies and ultimately, the clinic. And on Slide 9, we show the consequences of developing those integrated technologies. These technologies reduce the time and cost of the identification of such development candidate molecules as well as yields molecules with enhanced quality versus traditional efforts. We have here the time lines associated with traditional drug discovery, it typically requires synthesis of around 5,000 molecules and 4 to 6 years of work to identify a molecule that can be advanced into clinic. Using these technologies, we're able to cut that time line in half by evaluating billions of molecules. Now routinely, we can often do tens of billions, hundreds of billions of molecules to find a molecule with more optimal properties that can then be advanced into IND-enabling studies in clinic. And with that, I'm going to hand the presentation over to Karen Akinsanya, who will highlight how we've been using these technologies to advance our internal drug discovery efforts. Please, Karen.

Thanks, Robert. And so as you've heard from Robert, we have been able to deploy these approaches over the course of the last couple of decades, and I'll explain some of the outcomes in a moment. Our strategy is really multifold to continue advancing the technology and also applying that to small molecule and biologics discovery to collaborate in order to both work on the most interesting projects but also to maximize the value proposition for the IP generated through the platform. In addition, I'll be speaking about our internal programs. Our goal here is to advance those wholly-owned programs into development ourselves or in partnership in order to maximize the opportunity for those products and programs for commercial and we hope, patient success. On Slide 12, what you can see here as I alluded to, over the last decade or more, we've been deploying this technology in collaboration with partners. And you can see here, going back to the programs we worked on with Agios, which are now FDA approved, there are a number of programs that have evolved from the use of the platform that are now in the clinic in Phase I and II or are in IND-enabling studies. And what you can also detect here is that we work with a variety of companies. You can see here that there are companies that we were involved in co-founding like Nimbus and Morphic and a number of other small companies. But in addition, we work with large companies like Sanofi, Takeda as well on programs that they have an interest in accelerating through use of the technology. This slide also represents a very broad array of target types, some very challenging targets in here that benefited from the use of the platform. On Slide -- I'm sorry, it's 13, you can see here the wholly owned pipeline and our collaboration with BMS and Zai Lab. So these are programs that our drug discovery group that's now about 100-or-so people are pursuing. This includes the 3 wholly owned advanced projects in discovery, MALT1, CDC7 and WEE1. MALT1 is now in IND-enabling studies and as well as the 5 programs we're working on with BMS and a new program that we recently announced working with Zai Lab. I'm going to take a moment or 2 to speak about some of our programs at a very high level. So on Slide 15, you can see a description of the MALT1 program. This is a target that is in the BTK NF-kB pathway. Essentially, this target is involved in signaling of BMT cells and impacting a diverse large B-cell lymphoma. It's been shown that MALT1 fusions are responsible for the proliferative capacity of those cells. And indeed, we've been able to show, as have others, that when you inhibit MALT1, you can actually suppress NF-kB signaling. On Slide 16, we describe at a high level the application of the technology to this program. We have deployed the physics-based methods exploring billions of compounds. In this case, 8.2 billion compounds were triaged in a structure-based drug design project and using a combination, as Robert described, of our physics-based methods, FEP, in combination with machine learning and deep learning methods, we were able to conduct both the identification of novel compounds but also the optimization of those molecules through what's called multiparameter optimization, where you're balancing a lot of the properties that Robert described. We're on track, actually as we said to move this program forward. After synthesizing only 78 molecules, we found development candidate-like material and those compounds, as I mentioned, are now in IND-enabling studies. I won't go through this in any great detail. But just to say that the compounds coming out of these calculations obviously go in vivo. And what you can see here is, we've demonstrated antitumor activity in combination with ibrutinib and other BTK inhibitors. We've also shown very nice target engagement for this target, on Slide 17. Moving forward to Slide 19, another example from our pipeline is the CDC7 inhibitor program. Here, we were pursuing a design challenge that has been in existence over the last 20 years for this target.The goal being to balance exquisitely potent CDC7 inhibitors with great drug like properties, previous generations had issues with PK and selectivity. So the goal here was to come up with a picomolar inhibitor. On Slide 20, again, I won't go into a lot of detail, but just to say that not only have we found these inhibitors for CDC7 are picomolar. In fact, we believe these are some of the most potent kinase inhibitors identified to date. These combine quite well with BCL2 inhibitors, PARP inhibitors and ATR inhibitors as well as WEE1 inhibitors, showing very nice synergy in vitro in a variety of cell lines. And on Slide 21, this translates into in vivo activity. You can see in the colon cancer model, we see a very nice inhibition, indeed regression of the tumor growth. And also, we've shown very nice target engagement preclinically, which we plan to translate into the clinic in the near future. We've also looked at AML, which shows, again, very nice effect of the CDC7 inhibitors from our program on a well-characterized cell line, which is both shown in vitro but also here in a xenograft model. And finally, I'll end with WEE1 on Slide 24. This is a very interesting target that has been recently shown to have monotherapy efficacy data in uterine serous carcinoma as well as ovarian carcinoma. You can see the data presented here from the AstraZeneca program that was revealed at ASCO 2020. We also believe, of course, that WEE1 inhibitors will work well in combination with other agents. And that's important because that was the key design challenge here, to come up with exquisitely selective WEE1 inhibitors but also avoid some of the issues seen with prior inhibitors of WEE1, including CYP3A4 time dependent inhibition as well as off-target activity that make it difficult to tolerate the drug. We employed a unique feature of our platform, which is protein FEP. It's a method to actually simulate the impact of changing amino acids in the target itself. We did this to really understand the rules that govern selectivity around WEE1. And without going into too much detail, you can see the early compounds and competitor compounds that hit a number of kinases using this high ultra-scale method to model the effect of changes in amino acids, we were able to come up with a number of unique compounds that are exquisitely potent on WEE1, but are also is very selective. So Slide 26, a just to say that the molecules that we are now characterizing are indeed very selective, very potent but also have been able to dial out the CYP3A4 TDI liability, and we believe this is going to be an important molecule, set of molecules for patients and physicians and payers to be able to maximally -- obtain maximum potential of the WEE1 mechanism. So we're rapidly progressing our pipeline, as I mentioned, with programs now moving into IND-enabling studies, and we expect our first IND submission in the first half of 2022. We're also expanding our pipeline as those programs progress with new programs in additional disease areas starting this year. With that, I'll pass the podium back to Robert.

Thank you, Karen. So we wanted to summarize with Slide 28, our view that the future of computationally-driven drug discovery truly is exceptionally bright. But on Slide 28, you see that computer performance continues to increase at an exponential rate and the availability of such computing power with cloud computing resources is growing ever easier, which allows these types of methods to be applied at an ever broader scale. Complementing that increase in computing power, the percent of the human genome that is characterized with high-quality structural information of the type that is needed to apply these types of assets with high accuracy is also growing, which means more and more drug discovery projects will be amenable to this type of approach. Likewise, as there are more structures available in greater computing power, the number of compounds we can explore computationally define those few rare molecules that have the property balance associated to be advanced into clinic also is being increased where we imagine in the near future, we'll be able to explore trillions of compounds per program. And then likewise, we're continuing to make great strides on the long-standing scientific goals of computing various properties of extremely high importance to drug discovery and materials discovery with very high accuracy. And we imagine in the coming years, we'll be able to compute as many as 10 such properties with such very high accuracy in detail. That means in the future, we would expect the discovery of high-quality development candidates within 1 year of program launch to become essentially a routine outcome, and we're very excited to play a part and bringing about that reality. Slide 29, we wanted to speak to our company growth strategy, which is to invest broadly in all these lines of business. We are continuing to develop and improve our computational platform, which supports all of these various activities. We are continuing to work with pharmaceutical companies and biotechnology companies to adopt our platform at ever larger scales. We are continuing to deliver on our collaborative drug discovery programs as well as to initiate new collaborative drug discovery programs with partners. We are investing heavily in our material science software where the same underlying physics that leads to molecule binding to a protein often can be related to all sorts of different materials questions. For instance, the same thing that makes a drug molecule soluble is the same thing that makes materials additive, soluble or not soluble. We're working with material companies to adopt our computational platform and working with them to enable them to see similar results in their materials, discovery efforts. More recently, we've initiated an internal drug discovery pipeline that Karen has wonderfully summarized in the previous presentations -- actually, the previous part of this presentation and plan to initiate new programs connected with that internal pipeline. And then lastly, in the same way, we've been able to establish a highly successful drug discovery collaborations, we plan to establish similar material science collaborations. And with that, thank you so much for your attention. We're all happy to answer questions.

Great. Thank you. Thanks, Robert, and Karen. That was a great overview. Really appreciate that. I guess first 1 for me, I think you just touched on this, Robert, but I want to dig into it a little bit deeper, would be some of the topics you talked about on Slide 28. And I think one of the biggest questions we get is Schrödinger IPO-ed about 1.5 years ago, but the company has been around for decades. So sort of why now, what's come together? And you ran a lot of it on this slide, but it would be great if you could go into a little bit more detail in terms of the underlying technologies, both on the computational side and on the biology side, things like cryo-EM and just sort of how do you see the entire ecosystem the way it looks today versus what it was 5 years ago, 10 years ago, 20 years ago and sort of how that translates to the value of the platform that the technology generates?

Yes. It's a great question. There's really 4 major catalytic trends that have all come together to affect the types of outcomes we're now seeing. The first major trend is compute availability, it's just night and day what it was back in the '90s, 2000s. The advent of GPU programming, for example, was hugely transformative in terms of being able to apply these types of physics-based methods at very large scale. And then when those GPU technologies became available at a very large scale with cloud computing resources, one could first utilize these methods in a way that would have been cost prohibitive prior to that type of compute technology being so broadly available. The second complementary trend were advances in cryo-EM, x-ray crystallography. And then also recently, de novo structure prediction has been making great strides and then many might be familiar, for example, like the AlphaFold technologies, where more and more high-quality structures are available. And then once you have that high-quality structure, it becomes a wall post problem, okay, what is the molecule that will therapeutically bind to the high-quality structure of this target and then affect a therapeutic response. So then complementary to those 2 external catalytic trends are 2 that Schrödinger really has been spearheading. The first trend Schrödinger's been spearheading is the development of highly reliable, broadly applicable free energy calculation technology, small molecule force fields, different technologies that allow for these molecular properties to be calculated with very high accuracy that benefit from all of this available computing power and high-quality structures as well as integrate those technologies with machine learning methods that allow them to be applied at very large scale, allowing currently hundreds of billions, but in the near future, trillions of molecules to be evaluated for multi-parameter optimization. And then the last major investment we've been making, if you go all the way back to I believe it's Slide 8 is to integrate these technologies with enterprise informatics and scalable cloud systems that allows for all of that in silico calculated data and the experimental data to be available to project teams at the same time to facilitate fundamentally better decisions that would be possible in the absence of these technologies. So it's really all of those things coming together in a highly usable way for project teams that have enabled these outcomes.

That's really helpful. And then you mentioned some of the platforms you're developing. Another key question we get is, how hard would this be for someone else to duplicate outside of the field. If you -- if Google throws enough money at it and enough expertise at it, is this something that can be developed somewhere else? Sort of how much of this really is proprietary to Schrödinger? And how long would it take for the -- for these solutions to be developed elsewhere?

Yes. So we keep a careful watch on what other parties are doing, obviously. That includes industrial parties as well as academic parties that are working in this space. In the area of high accuracy physics-based calculation of properties of relevance to drug discovery, we believe our advantage in the atomistic methods would be something like you'd be a committed effort of something like 5 years for someone to catch up. If we decided to stop further investing, which obviously we're not going to do, we're going to continue to improve these methods, further build out the performance characteristics and broaden their applicability to more and more high-value endpoints. So I'll admit, I tend not to think about it in terms of time to catch up as much as I think about what we can be doing internally to even further accelerate our lead in terms of what's possible. And Joel, I see you turned on your camera. Anything you want to add there?

No. That's exactly right. We're investing heavily in the underlying platform on an ongoing basis. And you heard from Robert the elements that we're trying to refine and add and the capabilities that we intend to introduce over the coming years, we think we'll keep the moat that we have today.

And you've added new capabilities over time, right? I mean it hasn't been -- I think that's another point but it's not one solution, the Schrödinger suite is a number of different applications, each meant to target a different part of the drug discovery challenge or sort of process. Robert, maybe or Joel, could you highlight a few that you think are more notable that have come online in recent years? I mean, for example, FEP+ is something that we've heard a lot of talk about. And obviously, that's a relatively more recent addition. So could you talk about the evolution of the platform or the solutions as you keep sort of innovating and adding more capabilities?

Yes. So this is a place where the synergies or business models really become very apparent that are pursuing collaborative and internal drug discovery projects at that same time, we're developing software for use throughout the pharmaceutical industry and biotechnology industry. Means that we keep a really -- what's the right word, good track of where we could invest to be very high yield in terms of the value provided both to our software customers as well as the internal and collaborative programs. So for example, we would like to expand the platform to enable targets with more limited structural biology support, where there might be a few or no earlier tool compounds or known small molecule inhibitors and there has been really great strides made recently to be able to compute, for example, protein reorganization upon ligand binding, which is a key property that has to be modeled for some systems or hit discovery technologies that have gone through really tremendous improvements and then also things like breadth of applicability ways to optimize the potency permeability balance that might be very important to get a molecule with a particular bioavailability for a particular system. So these are cases where I want to be kind of mindful. I don't want to go like too far deep into the details, but there's a wide variety of accuracy improvements, breadth of applicability improvements and sort of what I'd call sort of system amenability improvements that are allowing us to accelerate more types of drug discovery projects more effectively.

Got it. That's really helpful. And I think you mentioned sort of the various parts of the business model, and that was exactly where I was going to go next was sort of the interplay between the software business and the collaboration business and the internal pipeline. So I guess next question, let me pivot to Karen a little bit. The evolution of the internal pipeline. How do you see that investment balancing going forward and that focus of the business going forward? And I guess a bigger picture question. When we first started interacting with Schrödinger, we thought of you very much as a platform company, as a technology company. Now it's sort of in this hybrid period, if we look forward 5, 10, 20 years ago, is this predominantly a biotech company? Sort of how do you see yourself evolving over time from a business model, from a commercialization point of view?

Well, I think the first thing to acknowledge is what Robert was alluding to, which is with every drug discovery project we take on, whether it's one of our own or what we're doing collaboratively, we're actually very focused on solving design challenges. And what that means is that with each and every program that we run, we learn new things. When I spoke about the WEE1 program where we were able to essentially understand the rules that govern selectivity, and this is kind of a breakthrough in the technology. We kind of call it the ability to do mutagenesis of proteins in silico. Now I imagine if you were able to deploy that more broadly across a number of programs, target classes. So the opportunity for drug discovery, we think, is very large in terms of being able to execute on programs, but also to be able to learn how to solve some of these big challenges. And because of that, the ability to run both drug discovery on standard and challenging programs, but also to keep adding these improvements and these breakthroughs to the platform, in our opinion, means that this is a sort of virtuous cycle that we have to keep going. We want to be able to solve very new design challenges, work on different modalities of targets. Some of that work is already beginning. The idea of working on different areas like biologics and protechs, et cetera, which are very exciting, and we think will be here to stay. So combining our drug discovery efforts with the platform development is really on plan. I will also say that, of course, our programs are going to continue to advance. And that means that the opportunity to derive value from those programs will also increase over time, not just in terms of the individual programs, but also the breadth and number of programs that we'll be working on. So the pipeline will grow both in maturity, but also in size. And I think that does obviously offer a very interesting opportunity to derive value from the IP that we're generating, not just collaboratively, but in a wholly-owned fashion. I don't know if Joel wants to add anything to that. But I mean some of the terms of our collaborative deals mean that as these programs transition they're going to be associated with a significant opportunity to add revenue to the pipeline of already existing software revenue.

Sure, sure. Yes, I'd just add that if you look at -- to get to your question directly, Mike, we believe there is a tremendous opportunity for growth in the software business, in the collaboration business, the groundwork that we've already established there with, for instance, a BMS deal and also our internal pipeline where we're creating, we believe, a lot of value there. So we have found this formula for a balanced business model where we're able to leverage across all of those activities, the underlying technology and the differentiation that it provides us and the competitive advantage that it provides us across all those business -- businesses. And so we're investing in all of those areas. And as you heard -- as Karen mentioned, there is a high degree of synergy across them. So we see a future where all areas of our business are growing and are a very significant contributor to value for the company.

Okay. I want to remind investors that if you've got questions, feel free to send them in via the Veracast portal or you can hit us a bump over chat or via e-mail, but don't hesitate to throw the questions in. Right along that line, though, Joel, I want to continue on that point and what Karen was touching about was sort of the interplay between the various segments. Have you run into any issues where customers are a little bit reticent about partners kits, if they worry that -- are they working with someone that could be a competitor where if you're developing your own CDC7 or MALT1 program, for example, they may not want to collaborate with you on something where they fear that there could be data breach and IP breach because you've got this very dynamic where you're providing them with software, you're also potentially collaborating with them and you've also got your internal pipeline. So are there any risks there or any concerns there in terms of what role you're planning at any given point in time?

Yes. We have the ability -- we have been able to avoid that, and we have a firewall between the 2 sides of the business to prevent key data from going over to their side, and we haven't seen a problem in that regard. I don't know I'll let Karen and Robert maybe speak further on that, if you have anything to add.

Yes. I mean I'll just say that as you said, Joel, there's a strict firewall. All of the customer data is completely segregated from any data that we might be generating on our programs. In fact, further than segregation of data, I think Robert can add to what it is we actually do see from customers.

Yes. The firewall protects both sides of the business and then we have procedural and technology-related solutions to make sure that there is no confidential information from software customers, everyone moves over to the collaborative and internal drug discovery side of the business and vice versa. And we've -- when thoughtful customers or collaborators have questions about it, we provide them the information they need so that they have the comfort and confidence to move forward with us. Those are very important conversations, obviously, but they haven't been problematic really. So we're excited. We've been able to establish that firewall and be able to continue to pursue both lines of business in a way that are beneficial and complementary with each other.

The other thing I might add, Mike, is that, in fact, it's quite the opposite. But we think that pharma and biotech, in general, are interested in this incredibly large pool of potential targets for therapeutic benefit for patients. And in fact, I think our collaboration's a testament to that. But there's a lot of targets to work on. And actually, whether they're using the software internally or whether it's collaborating with us on the cutting edge of how to apply the technology. There's a lot of inbound interest actually in working with us to accelerate drug discovery.

It's Derik. I want to jump in, if I may. Obviously, we've paid a lot of attention to AlphaFold and some of the de novo structured prediction programs. And it sort of has driven a question of is there some potential to make demand for cryo-EM less, right, and for [ NMR ] less and for X-ray less. I'm sort of thinking about this is how do you sort of see -- there's always -- is there a fear that these other -- the physical methods for actually [indiscernible] are going to become less useful? Or do you need those to sort of like to validate the AlphaFold data?

Dan (sic) [ Derik ], so keep in mind, AlphaFold right now as it currently exists cannot be used to attempt to model, for example, ligand-induced confirmational rearrangement where we know for a great many protein systems when a ligand binds that induces all sorts of important structural changes to approach. Functional activity often has a structural basis where when a ligand binds the bridging will reorganize in a way that affects a certain function. So I more view cryo-EM, x-ray, de novo prediction and then also other orthogonal methods of structure optimization refinement to be ultimately complementary that the goal here is to elucidate systems, and one wants to have as complete an armory as is possible to facilitate the elucidation of those systems and discovery efforts for those systems. So it's really the utilization of these technologies and in an integrated way, the use of these technologies together that can be used to accelerate discovery. So I don't -- I personally at least don't see any reason to think that AlphaFold has made cryo-EM instrumentation, for example, suddenly of 0 value. We believe -- we don't see evidence of that, maybe say it that way.

In fact, I mean, I think over the last decade or so, the growth of access to x-ray and cryo-EM structures has been exponential. And 1 application actually of the protein structure predictions from AlphaFold is potentially an even faster growth in access to structures -- experimental structures because it allows one to potentially inform construct design. So we see them, as Robert said, being very complementary and synergistic actually as 1 goes to validate and understand these protein structures.

Great. I had a couple of questions coming from clients, but I want to make sure I get 1 more in before we switch those just because I know it's a common point of discussion, Joel, regarding the software business sort of the outlook for the year, 10% to 19% growth. There's a lot of seasonality involved here in terms of expectations for the first half and the first 3 quarters versus what's expected for the fourth quarter. So could you again sort of walk us through on the software side of things, how are these contracts structured? How are they paid? How much visibility do you have? When do customers enter the so-called negotiation period? Just walk us through sort of the actual financials there and how the business operates in that sense?

Sure. Thanks, Mike. So we do have seasonality in our business. We historically have seen a large proportion of our full year billings occur in the fourth quarter. That's been the case in the last couple of years, and we expect that to be the case this year as we've talked about. The revenue from that surge in billings at the end of the year results in -- tends to result in larger revenue in the fourth quarter than the previous 2 quarters generally. And also the first quarter, some of that revenue gets recognized in the first quarter, depending on when the actual contract starts. And that's just a historical -- that's when we established a lot of our large customer contracts decades ago and they come up for renewal. So when we look out at the full year, we have really good transparency on the base business, I would say. And what I mean by that is when you look out at all the 4 quarters of the year, and we know all the contracts that are up for renewal and when they're up for renewal, what kind of contract it is, is it hosted, is it on-premise. And we also are having discussions with these customers, many of whom have been with us for 15 years or more, certainly our most -- our largest customers. And we are continually talking to them about the benefits that Robert described about large-scale application of our processes that we're seeing benefits on -- with our internal programs, astounding really profound benefits. And so we saw last year, obviously, big surge in upsizing from a lot of customers. We saw the continual addition of new customers, including, very importantly, on our material science business, which is still early in its life cycle. And those decisions to upsize don't happen in a uniform way year-to-year. We've guided to and talked about growth this year that will be more in line with that will -- that reflects the fact that some of those large customers had upsized last year need to see results from application of their new level of deployment. And that sometimes that takes more than a year. So that's the part that we -- that can be variable heading into the fourth quarter. We're having discussions about what level customers should be aspiring to. We view that as a multiyear process. I think last year very much showed us what we're capable of. It's important to note that we believe that we're -- that customers, even our largest customers, while we know, even our largest customers are deploying our solutions at 10 to 100x lower throughput than we are deploying it on our internal programs. And so we believe the -- even after a big surge last year and continued growth this year that there is a tremendous growth opportunity ahead. And so we're working over multiyear plan to get our customers to those levels, and we see great underlying momentum. Mike, you're on mute now.

Yes, I am. We're almost out of time. I appreciate the color there, but I want to squeeze in 1 more sort of as a concluding question. We've had a lot of companies present today. We've got more in the afternoon, and you can sort of tell what the entire theme of the day is about. But probably one of the most common questions we get is sort of how do you differentiate all these platforms. So if you could really sort of drill it down to a few key points. How do you differentiate Schrödinger versus some of the other companies we're hearing about today about Recursion and insitro and Atomwise and sort of what's unique about the platform? What's unique about your approach? Where do you fit into this broader ecosystem?

Yes. So one of the key differentiators is the quality, accuracy and breadth of applicability of the physics-based predictive technologies that underlie the platform. The other dimension of differentiation is the degree to which we've integrated those highly predictive physics-based technologies, with other supportive technologies, such as modern machine learning technologies to greatly increase the addressable scale of the platform and the way we've integrated it with enterprise informatics technologies that allow for all the experimental data and all of the computer data to be available by all members of the project team at the same time to facilitate better decision-making. So in our view, those are several dimensions where our computational platform really is unique versus other things we're aware of.

Yes. And I would just add quickly that a lot of companies are leveraging machine learning in lots of different ways that affect the drug discovery process. But we're really focused on design of medicines, ultimately, there's 30 programs now that have made it through discovery or into Phase I and II or even onto the market. And we think that focus on design is what differentiates us perhaps from those who are using image analysis or other approaches to characterize biology. We think that's important. In fact, we leverage that ourselves, but it's different from our platform.

Great color. Great. Thanks so much, Karen and Robert, I think that actually summarized it perfectly. I really appreciate those distinctions. So thank you so much. With that, we're out of time. So I want to thank all 3 of you for joining us today. Investors, I want to thank you for listening in, and I hope you found the session useful and look forward to speaking with you soon.

Thank you.

Thank you.