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

Good morning, everyone. I'm Jaren Madden, I'm the Senior Vice President of Investor Relations and Corporate Affairs here at Schrödinger. I'd like to welcome everyone who's in person with us today as well as everyone who is listening in virtually to Schrödinger's first Platform Day. Before we begin, I'd like to share some really exciting news. Today, Newsweek issued its 2022 list of America's top 100 most loved companies and Schrödinger is #21 on the list. I'm really proud of the company and the culture that employees have worked hard to build, and I'm pleased to be highlighting our progress with you today. Also a logistical note for our webcast participants. There's a box in the webcast window where you can ask a question. I will collate submitted questions. We'll respond to as many of them as we can during our Q&A at the end of this formal remarks. If your question is not answered, please feel free to reach out to ir@schrodinger.com. Now a few other logistical notes. Before we begin the formal presentation, please note that during today's discussion, our team will make statements that are forward looking and made pursuant to the safe harbor provisions of the Private Securities Reform Act of 1995, including, without limitation, statements related to our computational platform, software and licensing business as well as our collaborative partner and proprietary programs, including potentially favorable profiles of the molecules as well as the timing of IND submissions and the initiation of clinical trials. These forward-looking statements reflect our current views about our plans, intentions, expectations, strategies and prospects, which are based on information currently available to us and on assumptions we have made. Our actual results may differ materially from what we project today due to a number of important factors, including those described in our SEC filings, including our most recent 10-Q that we filed in August. These statements represent our views only as of today. We caution you we may not update them in the future. And with those logistics out of the way, I'd like to turn the rest of the program over to Ramy Farid, our CEO; and the rest of the Schrödinger team.

Thanks a lot, Jaren. Welcome, everybody. We're really excited to see you all here. Welcome also to everybody who's on the webcast. We're really excited today to talk to you about the company. Our focus, of course, as the name of the event implies will be on our platform. And we're -- There's a lot of excitement around the role that computation plays in nearly every aspect of our lives. And in particular, in drug discovery and in materials design. But of course, with all that excitement, often comes a lot of confusion and hype. I mean that's the kind of thing that inevitably happens and we think really you should demand from any company that's claiming to have developed a transformative platform that they actually talk about the platform. What it does? What it's ... How it works? And also of course, provide validation of the platform? Does it really do what the company is claiming to be doing. So we are going to tell you our goal today. Is to tell you about our platform, but most importantly, and you're going to see a lot of the focus is going to be on validation of the platform. So that's going to be where we're focused. So maybe we can advance to the slide that has the agenda and the people that I am I'm the one doing that... Sorry. Awesome. Sorry about that. I forgot that I had control over it. Of course, obviously, we're going to talk about the platform, but we're also going to tell you about business outlook and opportunities you'll hear. So let me tell you who you're going to hear from. So after I give a few sort of opening remarks, you hear from Robert Abel, who's our Chief computational scientist. You'll also hear from Karen Akinsanya, who's our President of R&D Therapeutics. You'll also hear from Hamish Wright, our VP of Translational Science; and finally, from Jeff Porges, our Chief Financial Officer. So hopefully, it's clear what the goal of this platform day is. Let me start -- Before I get into sort of talking about the platform, let me just give you a few slides and sort of overview of the company. I think you're all pretty familiar with this, but I think it's good to sort of set the stage. So of course, we've developed a computational platform, a technology platform that you will see today, I think we will show very clearly, is transforming drug discovery and materials design. We're going to focus on drug discovery, though, today. And we're going to show you that the platform is delivering on the promise. That of delivering drugs and drug candidates faster -- Higher quality drugs faster at lower cost. And you're going to see some really clear data to demonstrate that we've also improved the probability of success compared to traditional methods. As you know, we licensed our software to pharmaceutical companies, biotech companies, industrial companies, materials companies and academic labs worldwide. So again, both in life sciences and materials science. We also as you'll hear about today, form collaborations on the drug design area, and again, that's going to be the focus today, but also in materials design. You'll also hear about today, our proprietary pipeline. You'll hear about programs that are at the discovery stage and the development stage. And in some cases, those programs are wholly owned and in some cases, they're already partnered. So one more slide sort of just give you a sense of the company, and I'm going to be talking a little bit about this. We have a long track record, 30-plus years of scientific innovation and computational chemistry and around 800 employees, a little over 800 employees. And you can see a very large fraction of these employees have advanced degrees, PhDs. And half the company, I mean this is a real serious commitment to advancing R&D, half the company is dedicated to research and development. And as you can imagine, these technologies are pretty advanced. And so we've made a really serious commitment into supporting our customers, educating our customers, educating both at the sort of really early stages, even students, but of course, also our customers. So there's a really serious commitment to supporting education around these advanced technologies. And this is a little bit of a detail, but we do have quite a few releases every year. We have 4 releases a year, where we introduced new functionality and improvements to the software. Again, you're going to hear a lot about that today. So before we get into the details of talking about the platform, I think it's useful to give a little bit of an overview about what it is that we're doing. Again, I started off by saying there's a lot of confusion. What space are we really in? And you've heard a lot of different things about drug discovery. And we all know, of course, that it's -- you keep hearing, yes, it's hard and it's prone to failure and it's expensive. But it's not so clear actually why that is, really. So I'm just going to spend a little bit of time explaining why designing drugs is so unbelievably hard. Why it takes so long? Why it's so expensive? Again, you keep hearing about this, but most importantly, why it's subject to such high failure rates. The reason, the primary reason is that the goal of drug discovery is to identify a molecule that possesses a whole bunch of properties. You see them listed here. A molecule to be a drug obviously has to be potent. It has to bind to the target that you're targeting. It has to be selective. In other words, it can't bind any other-- Really shouldn't bind any other proteins because, of course, that can cause adverse effects. In order for it to actually be a drug, it has to be soluble in water, obviously. It has to be bioavailable. Obviously, if not bioavailable, it isn't going to be a drug. It's not going to do what it's supposed to do, and so on, you can see the list, Permeable and so on. Now here's the challenge, though, and here's the problem. Physics is sort of not on our side in this particular case. These properties are all fighting each other. That is when you generally, when you have a molecule that's potent, it's not soluble. That's just physics. There's not much we can do about that. And then when you have a molecule that's made more soluble, it won't be permeable, and won't get through cell membrane. And if it doesn't get through cell membrane, it's not a drug. All of these properties, they're complicated in and of themselves, but they're fighting each other. This is why you may often hear that drug discovery is like a whack-a-mole problem. And you're going to see -- let me show you now the consequence of the fact that these properties are anti-correlated. So here's the kind of representation of a typical drug discovery project. So on the first column is, let's say, a molecule, you see the representation of the molecule, first column, okay? And that molecule, let's say, came from a 6-month effort or 1-year effort and high throughput screening, you screen hundreds of thousands of molecules, you finally found a molecule that's potent. And now you need to make it into a drug, which means you need to start making changes to it so that it actually has all those other properties. So here's how a typical project will go. You'll make a change to the molecule because you're trying to make it selective. You're trying to make it not bind all these other proteins in the human body. So you make one atom change. That's what that blue atom there is in the second column. And you're excited, you made it selective. But invariably, what happens is now the compounds is not potent anymore. So now you make another change. That's the third column. And okay, now you're getting excited. You've got a potent molecule and you have a selective molecule. Now let's try and fix solubility, the next property. So you make another change. And you see, okay, I made the molecule, it's not that hard actually to make a molecule soluble. You're going to have charge, you have some polar atoms. It's not that hard. But again, invariably, what happens is you've now made the molecule not potent. And then you keep going. You see what happens? The fourth molecule. And by the way, I'm showing it as one molecule. Usually, hundreds of molecules are made to try and solve but I just couldn't represent that. You keep making molecules. You see what's happening here. It's kind of very representative of a typical drug discovery project. The fifth molecule, you finally identify a molecule that's both potent and soluble, but now it's not selective. And now look, if you look at this example, the last mole is back. You're back where you were in the third step. That's what we mean when we say whack-a-mole. And this really happens. And so here's the result of this. Thousands of molecules get made thousands and even sometimes tens of thousands of molecules get made in the lab, that takes 5 to 10 years to do. Tens of millions of dollars. And of course, you can imagine, given how hard this problem is and this anti-correlated nature of these properties, large majority of programs that start with that potent molecule actually don't achieve the end goal of filing an IND. You see 2/3 of programs fail to deliver an IND and it's because of this whack-a-mole problem. What's our vision for the future? What vision for how this could be? Well, now it should start to seem kind of obvious. If we can do 2 things. One is compute the properties of molecules, all of them with high accuracy, and we can do it on a large enough scale where we can test -- let's not get into the numbers right now, test a very, very large number. Remember, you're looking for a needle in a haystack, so to speak, sorry to use that cliché. That's the whole idea. It's very, very hard to find a molecule that meets all these properties. So what do you have to do? If you can compute all the properties, you can obviously do it on a very large scale. And then you can identify that one sort of magical molecule that somehow balances all of these anti-correlated properties. So that's what our focus is on. Our focus is on developing methods that can predict these properties and do it on a large enough scale so we can solve this multiparameter optimization problem or this whack-a-mole problem. How are we going to do that? At a very high level, there are basically 2 approaches. One is machine learning, knowledge-based. You now always hear this being referred to as AI because it just sounds cooler to call it AI. Just it's machine learning. In computational chemistry, when people say AI, they mean machine learning. There's a lot of excitement about it because of what we have seen in our -- almost in our everyday lives. I don't know how many of you are playing chess and go and your everyday lives, but you get the idea, you hear a lot about this. About these extraordinary things that AI is doing in playing games and image processing and obviously, in self-driving cars, some of you may even have some of that technology in your own car. And so the question is, can this idea be used to design drugs? I'm going to get to that. The other completely, totally different approach. And I can't emphasize that enough. It's not, in any way, knowledge-based machine learning, it's developing what we call rigorous first principles, physics-based methods. I'm going to show you on the next slide that this requires a really deep understanding of the underlying physics that governs these properties, potency, solubility, selectivity, permeability and so on. Again, I'm sorry to do this. I'm going to say the third time. These are totally different methods. And I'm going to show you in a second what I mean by that. Before I get into this, I think it's really helpful. I'm glad this is actually working. I'm going to show you a simulation of a molecule binding to a protein. I'll talk you through this in a second. But this should give you a sense of what we mean when we say things like understanding the physics that underlies these complex properties. What you're looking at, the green molecule there that's jiggling around is the drug candidate. Those things that look like birds are water molecules, of course. And what's happening is -- Sorry, in the purple surface is a representation of the protein. Underneath that surface, are hundreds of tens of thousands, hundreds of thousands of [ atoms ]. If I showed them all, you wouldn't be able to see anything. So we just represent the protein as a surface. And you can see the pocket there, and you see the molecules coming in and binding. This is what happens when a molecule binds to a protein. It comes out of water and binds the protein. These water molecules have to rearrange. The water molecules that were in the pocket of the protein have to obviously evacuate from the pocket, so that the molecule can fit in. And look, all that jiggling around, eventually the protein and the molecule have to adopt a certain shape. We call that a confirmation, to be able to actually accommodate each other. The physics behind this process is really complicated. I we're going to try and predict using first principles methods, let's say, in this case, binding of a molecule to a protein, it's pretty complicated. Again, we're going to talk about that. Now let me just spend a few minutes showing you pictures of cats. Now I'm going to talk about machine learning. This is a very useful way of doing that. What is machine learning? I think it's really important to understand that to figure out sort of how we're going to use these different methods? Machine learning, physics-based methods. So here's what machine learning is. I think you all know this, but again, I think it's useful to show this. With machine learning, you train a model. In this case, we're developing a machine learning model for detecting cats and photographs. So we obviously have to train it on a bunch of pictures of cats. If you have enough variation, a big enough training set of pictures of cats, now when you try and predict the weather the thing that you're looking at, which in the bottom there is another cat, that obviously isn't exactly like the ones above the model works pretty well. If you have a good enough representation of every type of cat. But what happens in these models when you show it another animal that hasn't been trained on. Obviously, the machine learning model knows nothing about dogs? I mean it's only been trained on cats. So it will probably mischaracterize this dog as a cat. That's the limit. That's the challenge of machine learning. It's developing a training set that's representative of the thing you're trying to predict. Now that's pretty clear, I think. Let's consider now applying this to design of molecules. I'm just going to show you, this is just a slide to just show you -- I'm about to show you a bunch of pictures of a molecule that's pretty complicated. But instead of showing the whole molecule every time, I'm going to represent the complicated part of the molecule with a yellow sphere and just focus on this group on the right. That's all the slide is just so you're familiar with the one about to show. Now let me advance. It looks like one of the molecules is missing on the right, but that's okay. I'm going to explain it to you. This is the training set. The training set is 2 molecules. One molecule has an atom, you see there, the green atom? The other molecule in this picture that's missing is another molecule looks a lot like that one. It just has a slightly bigger group than that green atom. Now, when you present it with, -- that's the training set, -- now when you present it with a new molecule, you see what the group there is, and you'll have to use your imagination here is somewhere between the size of the small atom, that green atom and the big group that's on the right there. This will probably work pretty well. This is interpolation. In other words, if you think about those 2 molecules, again, the one you can't see, the one you can see as 2 cats. And this kind of looks like a -- yes, that will work, and it will probably work pretty well. So the model can correctly predict that, that molecule is potent. Now, let's present it with a new molecule. So you tell me, does the molecule on the right look similar to the molecule on the training set? I'm pretty sure most of you would say, well, obviously, it's pretty similar. All I did which is move it down one position. It's the same -- actually the same atoms, just a slight variation. But it turns out that is a profoundly different molecule actually, even though they kind of look the same. And in fact, the machine learning model here would incorrectly predict that this molecule is potent. And the reason, by the way, goes back to that movie we just looked at. There are subtle differences in the water structure of these 2 molecules. The one with the molecule and that what's called the [ para position ] and when the atom is in this [ meta-position ]. These turn out to be completely different species. Now think about that. All we did was just move one at them. Now, if you think about the diversity of chemical space, you would need a training set that's way bigger than every species of animal on the planet to be able to capture this complexity. I hope that's clear. Almost every molecule that you make is actually a different species. It's called chemotype in chemistry. And so the training sets that are required to be able to capture all of this complexity are impossibly large. So it's not the same problem as image processing and, for example, detecting particular animal in a photograph. What you're looking at here is actually just a representation of the movie you just saw. It's the same thing. It's just a different way of looking at it. Here's what you have to be able to model when you're trying to use physics-based methods. You have to model the molecule in water, that's what's on the left. The protein in water, that's also on the left, on the bottom. You have to model what happens when these molecules adopt the confirmations that they're supposed to have when they come together? What happens when the water molecules leave? When these things bind, obviously, the waters that used to be around the molecule aren't there anymore, and what happens when they come together. You have to be able to understand every one of these steps in order to actually accurately predict these properties using first principles. We already established that alone machine learning models can't do this because you cannot build a training set big enough to capture all of this complexity. So by the way, the physics-based methods in this particular case actually accurately predict that this molecule is impotent, it has an IC50 2 micromolar. And I think that's a very exciting thing to think about. That this allows us to go beyond the limits of current knowledge, by having these physics-based methods that can solve these problems using first principles and not relying on what we already know. Now, there's no free lunch. These physics space methods, I'm going to focus on the right side of the slide. I just told you, you don't need a training set. That's pretty cool. That's exciting. And you can extrapolate into new chemical space. That's really exciting, and they're accurate. That's fantastic, but it comes at a cost, which is these are computationally expensive calculations. It's slow. Machine learning does have some very interesting advantages. There, as I showed, very effective at interpolation, they're fast. I didn't tell you that, but they're incredibly fast. And they can handle very large data sets. But as I just said, they require essentially, impossibly large training sets and obviously, you can't extrapolate. So this is a good point actually to hand it over to Robert. Robert is going to finish the story and tell you how we are integrating these 2 approaches to solve this sort of problem of each of them having sort of an issue. Okay. Is that a good?

All right. Thank you, Ramy. There's really 2 complementary approaches that have been historically pursued to predict the properties of drug-like molecules. Those are machine learning technologies and physics-based technologies, really very different, where Schrödinger has been breaking new ground is developing integrated approaches that combine physics-based simulation and machine learning methods to deploy methods to advance drug discovery projects that maintain the accuracy of physics-based simulations while having the computational efficiency and addressable scale of machine learning methods. Our ability to pursue this has really been driven by a long track record of scientific innovation at the company going all the way back to 1994, where we were able to figure out methods to speed up and maintain the accuracy of quantum mechanical methods such that they could be applied to problems of interest to drug discovery, that was followed on with advances related to molecular mechanics, protein structure refinement and eventually in 2012, a development of molecular mechanics for field. So the underlying technology that describes all of those molecular interactions Ramy was mentioning that has a comprehensive description of medicinally interesting chemical space. Once we were able to combine our advances related to force fields, molecular dynamics, molecular mechanics, we were then able to build out in 2016, a free energy calculation methodology that captures all of these things in a self-consistent way that could be used to accurately model potency, selectivity, solubility. And then in 2019, we were able to integrate this free energy calculation approach with modern machine learning techniques, in particular, active learning to allow the accuracy of those free energy calculations to be applied to very large sets of molecules. And then more recently, been able to extend this into problems related to protein refinements so that we could apply these technologies to ever more interesting targets of medical interest. So we're very proud of our publication track record. We publish all of these advances, so people can inspect the details of what we're doing. This is actually just a representative set of publications that have both been highly cited or particularly impactful. And again, we're very committed to our publication track record and would highlight here, for instance, our work on the glide docking program, which has been one of the most highly cited references in the history of the Journal of medicinal chemistry. We would highlight here the middle reference, which is our introduction of the first free energy calculation protocol that included a modern force field with broad coverage of medicinally interesting chemical space that allows for this accurate property prediction. And then more recently, work we've been doing to combine these free energy calculation methods with the addressable scale provided by cloud computing and integration of modern machine learning methods. So side by side with the scientific innovations, we've been building out the computational platform, which gives life to the scientific innovations and their ability to advance drug discovery projects. If we go all the way back to around the year 2000, the major focus was supporting virtual screening, which is a way to support hit identification, so that we can initiate small molecule drug discovery projects. But it was obvious we needed to do a lot more work. By 2010, we had broadly built out a whole suite of functionality that was allowing us to support not just hit identification, but also target validation and tractability assessment through improved protein refinement and drugability assessment of putative small molecule binding sites. And then by the modern day, we had built out a broad suite of functionality that was allowing us to support target validation efforts, hit identification and also really support lead optimization, which is where those whack-a-mole problems, Ramy's referencing, really come to the floor. Where we need to be able to accurately model potency, selectivity, solubility, permeability, these other properties to find that one molecule the project team is hunting for, which will allow them to select a development candidate molecule and move into preclinical development. We've also been excited that we've been finding opportunities to adapt these technologies to support preclinical development to expedite entry into the clinic for such molecules. I want to emphasize the deep integration we've been able to achieve between physics-based methods and machine learning methods. We have here a relatively simple example. Project teams have an enormous number of molecules. They could hypothetically synthesize. We can easily enumerate billions of such candidate molecules, potential molecules. We can't score $1 billion with full free energy calculations due to the computational cost Ramy referenced, but we can easily take 1,000, a representative sampling of that full space of $1 billion and overnight using cloud computing resources, compute 1,000 free energy calculations. And that creates a large virtual data set that we can then train a machine learning model to. So this will be a very approximate machine learning model, but it can be used to rerank that full set of 1 billion molecules to find the best 5,000 or so we can advance to free energy calculations. And then we have a rank-ordered list of some of the best molecules of that full set of $1 billion, and we can have the project team experimentally synthesized on the order of 10 molecules and through a really very extensive retrospective and prospective profiling, we would expect about 8 of those 10 molecules to materially advance the program along its intended endpoints. I wanted to give a sense of how this works in practice. So the experimental drug discovery project progression is described on the right side of the slide, and there's a computational cycle that's described on the left. We can use these technologies at every stage of the drug discovery process to evaluate billions of ideas. This facilitates very close interaction between the computational chemists and the medicinal chemist on the project, that through the enterprise informatics systems we're building out, have access to all the computational data and all the experimental data at the same time that facilitates better decision-making regarding what molecule should be advanced into the experimental drug discovery project progression, which allows for development candidate molecules to be identified more rapidly and with better properties than would likely happen otherwise. And that's really just summarized here where at the top of this slide and a traditional drug discovery paradigm where one has to basically use intuition, human judgment to try to guess the molecule one should synthesize. One typically makes thousands of molecules over 4 to 6 years to ultimately end up at a development candidate, which may have substantial property issues where one is forced by the realities of the pressure of the drug discovery project is to move forward with that development candidate. Using this type of computational analysis, we can support faster identification with experimental synthesis and characterization of fewer molecules, typically only hundreds, few hundred to identify drug development candidates with more optimal properties than would be achieved otherwise. And with that, I'm happy to turn it over to Karen and Hamish, who will present a variety of ways these technologies have been used in the context of active projects.

Thank you, Robert. So you've just heard from Robert, how the platform works. Ramy described the physics-based methods. I'm going to talk to you today about how these methods and this platform have been integrated into our collaborations and our proprietary pipeline. And just to give you a little bit of a sense of the description here, the collaborations of where we work our computational chemistry experts work with other companies on drug design and our proprietary pipeline, which I'll describe in more detail is where our internal team of [ crush ] functional experts, including computational chemists work to develop, discover and develop candidates for the clinic. So this is a long history. As Ramy said, we've obviously been developing the platform over a long period of time. But in addition, we've been establishing relationships for more than 15 years. And the collaborations really, as I said, were a computational chemists of working with outside companies. There's a variety of these types of relationships. Starting with Nimbus, which is a company that was cofounded by reading around 2010 and then a series of other companies that we have been involved in the founding of or at the inception of their pipelines. In addition, we have collaborations with Big Pharma, who obviously use our software and buy our software, but who work with us and our expert team on some of their programs. In addition, over the last 5 years, much more recent, we have developed an internal team that includes, of course, computational chemists but medicinal chemists, biologists, pharmacologists, toxicologists. And that group has been involved in the -- both identification of projects that would be appropriate for us to pursue both with our platform, but also from a therapeutic standpoint. And that portfolio of proprietary programs has been evolving over the last 5 years. Some of these are wholly owned and others, we've partnered with other companies. That includes the most recent transaction that you're aware of with Bristol Myers, where 5 of our programs were partnered with that company. Today, as part of the Platform Day, we announced a new relationship. This is with Lilly. This is a small molecule program. We have aligned on a target of interest. Schrödinger will be responsible in this relationship for the discovery and optimization of small molecules. In addition, Lilly will be participating in the program, participating in the use of animal models to characterize those compounds and will also be responsible for preclinical and clinical development and commercialization. Associated with this deal, the milestones you can see here, we're eligible to receive $425 million of discovery, development and commercial milestones for this program, and we have royalties ranging from low single digits to low double digits. The combined portfolio of collaborations, our proprietary programs and partnerships has continued to evolve. And we're excited by the number of programs that have transitioned from the discovery phase now into the clinic. And so what we're going to do today is give you just a few vignettes. We're not going to be able to go through too many of these case studies. But what you can see from this platform, this view of the collaboration programs is that we've worked on many different target types, many different therapeutic areas with a lot of companies. And indeed, 9 of those programs are already in the clinic. I've already talked about our proprietary drug discovery. As you're aware, we have a number of programs moving through discovery. And this year, we announced that our MALT1 program is the subject of an open IND, and we'll be talking about that as a case study. In addition, a number of programs have been partnered already, and you can see that on the right side of this chart. I do just want to point to the Zai Lab relationship. Our business development relationships have continued to evolve in terms of the type of structure. In the case of Zai Lab, we are eligible for milestones, but we are also able to participate in co-development and co-commercialization. We have that option at a certain point in the program. And our Lilly collaboration is now on this portfolio. So now we are going to switch to telling you a little bit more about some of the historical programs we've worked on that are now in the clinic, and we'll also talk about some of our discovery programs as well. I'll hand it over to Hamish.

Okay. So good morning, and I think we'll start the case studies off with a discussion of our novel ACC inhibitor program. This is a collaboration, as Karen mentioned, with Nimbus Therapeutics. So I think we're all aware of the public health tool that is imposed by NAFLD and NASH, where nonalcoholic fatty liver disease has a prevalence as high as 25% worldwide by some estimates. Now a settled CoA Carboxylase or ACC is the first step in the Fatty acid synthesis pathway. And it's also a rate-limiting step and for these reasons, it's been a drug target of interest to address NASH and associated metabolic disorders. Now inhibition of ACC 1 and ACC 2 results in reductions in fatty acid synthesis, tissue triglycerides on body fat and body weight and improvements in fatty acid oxidation and insulin sensitivity. And so in so doing ACC inhibition shows promise to address liver disease, type 2 diabetes and dyslipidemia. Now, historically, efforts to target ACC. We're focused on the carboxyltransferase or CT domain. Now the binding site here is highly lipophilic, and efforts here resulted in very low yields, so low numbers of molecules generated with very poor drug-like properties. So the design challenge was really to expand the search of relevant chemical space to discover novel chemotypes to address these limitations. In contrast to the CT domain, the Biotin Carboxylase, or BC domain has not been targeted before. However, the availability of crystal structure had a resolution of 2.3 angstroms with [ sorafin bound ], which is a natural product, suggested the availability of a novel binding site that could be druggable. And drugging this binding site would allow for disruption of the homodimerization of the ACC functional units and may be able to address both ACC 1 and ACC 2, which had been a historical challenge for this field. So using this co-crystal structure, scientists leveraged Schrödinger's water map technology, which reveals the location and energetics of water molecules in the binding pocket. So what you're seeing here on the left is that [ Sorafin ] does not displace the 2 water molecules, the 2 red spheres but suggested that this may provide an opportunity to capture increased potency if we could find chemical matter that could do that, they could displace these water molecules. Indeed, what you're seeing on the right-hand side is a potent ACC inhibitor displace [ disease ] unstable water molecules and capture additional potency. And so in summary, in this case study, water map was used to reveal an allosteric novel binding site in the BC domain. NDI-976 was discovered as a dual ACC 1/2 inhibitor, and this allowed for a maximal impact on the target and the reversal of lipid accumulation. Interestingly, NDI-976 interacts with proteins that are expressed in the liver, and this allows for a very favorable PK profile where the compound partitions to the liver. Gilead acquired the program in the second quarter of 2016 based on Phase I data that showed proof of mechanism and specifically on endpoints associated with de novo lipogenesis. And the compound was renamed GS 0976 or Firsocostat and is now progressing in a Phase IIB trial. At this point, I'd like to turn it over to Karen.

So moving on to another Nimbus example. The [ JAK TYK ] family, I think everybody is aware, has been a very productive family, important roles in psoriasis and a variety of other inflammatory diseases. In fact, there are 4 members of this family. They are complex in the sense that they are multi-domain. And I want to importantly point out that the kinase domain has been the focus of many of the marketed products that have come from this family. As you're aware, there are a number of approved JAK inhibitors. However, they are associated with some safety issues that involve heart function, clotting and thrombosis. And so the idea behind the TYK program was really to design an exquisitely selective TYK2 inhibitor. TYK2, as you know, has a lot of genetic support as many diseases that have been shown in humans to be associated with TYK mutations. Knowing what we know about the JAK inhibitors, the question was for the design challenge here, how can you design a TYK2 inhibitor that doesn't hit the JAKs. And in this case, the regulatory domain, also known as the JH2 domain, as shown in this light blue color was an opportunity to design allosteric inhibitors and avoid the kinase binding site. And this movie just describes the challenge of the kinase domain. So what you're looking at here is Tofacitinib bound to overlapping structures of TYK2 JAK 1, 2 and 3. And what you can see is that they are incredibly similar. It's going to be very difficult to design a compound that has unique properties for one of these unique bindings to one of these. And so this was the modelling challenge. And as I said, rather than focusing on the kinase domain, the focus was on the allosteric site. So this was a really challenging program for the reasons that we've just stated, and using free energy perturbations, which you've heard Ramy and Robert describe, we were able to score many, many thousands of compounds using FEP to identify molecules that were more selective for TYK2. But importantly, and I think this came through in the description so far, having a co-crystal of the compound bound to the protein and understanding the confirmational changes that are induced when that compound binds to the protein is absolutely critical. Ramy showed you a video of what happens when a compound binds to a protein. So having this co-crystal really opened up a new opportunity to understand the binding mode and the opportunities to optimize the chemical matter. And you can see that continued scoring of compounds once we had that co-crystal that was produced by colleagues at Nimbus. This allowed a further move up in terms of our ability to drive selectivity between TYK2 and JAK. And then finally, there was a breakthrough in the program. As you've heard, we can enumerate billions of compounds. And what that does for you is it gives you an opportunity to find very unique and diverse chemical matter. The collaborative team, the joint team identified a new subseries that basically took advantage of some conserved residues -- sorry, nonconserved residues between TYK2 and JAKs. And you can see that had a profound impact on the program. So after synthesizing under 400 molecules, we were able to identify compounds with about 560 fold selectivity. So let's wrap up this very brief vignette with a look at the clinical compound, also known as NDI-034858. You can see from this that the joint team was able to identify picomolar TYK2 compounds that had the desired effect in functional assays, IL-12 and other important assays, including interferon. But as was set out as a goal for the program, these compounds have very poor, in fact, almost no affinity for the JAKs and other off-targets, which are often the cause of safety issues in programs, as Ramy mentioned. So these highly potent and selective TYK2 inhibitors are being progressed by Nimbus. They have clinical data in humans showing good tolerability, also good target engagement. And this program is now in Phase IIB, as I'm sure you're all well aware, in moderate to severe psoriasis. I'll move on to another of our collaborators, Morphic. As you know, Morphic is focused on the integrins and the company was founded by Tim Springer, and we were involved very early on in this company. The focus of this particular design challenge was around alpha 4 beta 7, it's an integrin that's very important in the trafficking of cells, T cells from the circulation into the gut. And this is a validated mechanism. There is an antibody called ENTYVIO Vedolizumab is marketed and is helping patients, thousands of patients actually take ENTYVIO. The opportunity here was around the fact that this antibody is administered biweekly as an infusion and ultimately, bymonthly as an infusion. So the quest was to go after a small molecule inhibitor that could be taken easily at home. The challenge, though, was that there are some off-targets, in this case, after 4 beta 1 and its interaction with a different protein that potentially would have caused issues. So here, you can see that. The goal was to achieve potency of course. This is a fundamental property of all drugs, but also importantly, selectivity. As I just pointed out, the VCAM protein is bound by alpha 4 beta 1 and this is associated with a risk of PML, which is a rare but devastating form of brain infection. So the sort of goldilocks situation here was unique in that we were looking for potent molecules. They had to be selective to avoid this off-target, but they also had to be orally available. They have to have very unique PK properties. As you are aware, Morphic and Tim Springer, who's the founder of Morphic brought some very unique science to this field, and that's a hallmark actually of the companies we like to collaborate with. This was outstanding work to examine the binding mode of existing molecules. So there were attempts to go after small molecule inhibitors previously. But in this case here, which is this Roche molecule, it was pretty clear from this publication and the work that was done from the Springer lab and Morphic that existing compounds were potent actually against alpha 4 beta 7. But the structural explanation, structural biology explanation for why they also bind alpha 4 beta 1 was a breakthrough that was published in this paper. And it led to a very deep understanding of actually how to drive selectivity. Morphic produced over 100 proprietary crystal structures. And again, as I said, those confirmational changes every time you find a molecule are absolutely critical. And so looking forward then, with those unique insights, -- What was the goal on this slide? Well, you can see down in the corner a molecule called AGM. This was a benchmark. This was an early small molecule inhibitor that would have had to be administered a gram x3 a day. So it was sort of really available, very not the best profile. So that was kind of a benchmark. You can see that in the yellow star using FEP and a variety of methods, we were able to identify in the gold star potent molecules. But we're not just focused on potency as we've described this morning, we're also focused on other properties and balanced properties that make for an excellent drug candidate. And so on the y-axis, you're looking at permeability, and we've already described here that oral availability, when you're switching, we call it modality switch for an antibody to a small molecule is absolutely critical. And so the progression from red to green is actually how this program progressed. Not only did we find more and more potent molecules, but we were also able to use PKA simulations and modeling to drive compounds and that MDCK approximately 5, we needed to be above that line. And so using our platform, and you can see here, 8,500 free energy calculations to predict potency and also multiparameter optimization to balance properties, we were able to identify the candidate. That candidate is MORF-057. As you're aware, Morphic is progressing this molecule into the clinic. This is a fabulous collaboration between Schrödinger and the Morphic Sciences actually across a number of targets. This happens to be the most advanced -- you can see that the compound was potent and selective against alpha 4 B1 in the same way, actually, that the antibody Vedolizumab was. Unlike other compounds, I already mentioned the AGM molecule, not as potent and certainly not as selective. So the program is moving forward. Morphic is conducting a Phase IIA trial in patients with moderate to severe colitis and the Phase IIB study is planned for later this year. With that, I'll pass over to Hamish for the third case study.

It's a pleasure at this point to pivot to a case study starting with our proprietary programs. And in this case, we'll talk about cell division cycle 7 or CDC 7, which is a kinase that is a key cell cycle checkpoint for DNA repair. So starting in the [ contin ] on the left, demonstrating here the cell cycle, the transition from G1 to S-phase is a highly regulated event. And in S-Phase then DNA replication occurs. CDC7 has an important role in S-Phase in the initiation of DNA replication, also known as origin firing. And so in healthy cells, CDC7 activation resolves certain stresses that occur to repair DNA and allow the resumption of cell division. In contrast, cancer cells have a high level of DNA lesions, high replication for instability and high genomic instability, all of this leading to replication catastrophe that is the inability to successfully replicate. And so for instance, in the case of administration of PARP inhibitors, which cause trapping of the PARP DNA repair enzyme on the DNA, this calls a stalled DNA replication fork. CDC7 would typically be able to overcome this through its activity to protect and then restart those replication forks. Now in the presence of a CDC7 inhibitor, such as our development candidate, 2921, what we see is that the replication fork collapses and the DNA breaks, this leads to apoptosis and cancer cell death. So the design challenge in the context of the CDC7 inhibitor space has been one that's characterized by having compounds that really have not been very potent and also demonstrate very poor PK and poor selectivity. So if we were to go back sort of 20 years, the state-of-the-art there were compounds that were sort of double digit in terms of potency, double or triple digit in terms of potency and having poor PK and poor selectivity. Now if we fast forward 10 years from there, very modest gains. So only one order of magnitude had been identified or discovered in those molecules. And they were still showing relatively poor PK and poor selectivity. Now in contrast, Schrödinger's scientists took on this design challenge, we're able to discover picomolar inhibitors with drug-like properties that achieved our target product profile, all in a span of about 2 years. The cross-functional project team leveraged modeling at scale and deploy [ Schrödinger's ] technology to do this in order to drive very rapid decisions between chemistry, biology and DMPK scientists. Now we know the drug discovery requires a high degree of collaboration where functionary experts are constantly in communication and sharing data. And that's represented in the flow chart on the left-hand side of the slide. Schrödinger's CDC7 inhibitor project team leveraged a global collaboration tool live design in order to, in real time, share chemistry data, biology data, PK data and other data in order to drive this program forward very, very quickly and identify a development candidate in about 25 months. So how did they do this? Leveraging simultaneous modeling of potency, selectivity, solubility and permeability, we were able to achieve the target product profile. We'll start on the left-hand side, where novel type mining cores with favorable drug metabolism properties were identified. And those compounds were associated with a potency that was similar to what was being described in the literature. Schrödinger's scientists leveraged the platform and allowed -- and identified molecules had improved potency in the middle, but we're also able to simultaneously balance selectivity and solubility in order to come up with better molecules with better overall properties. And continued work with the technology allowed for the discovery of even more potent molecules and simultaneously improvements in permeability that allow the scientists to overcome a design challenge at that time, which were related to the in vitro-In vivo correlations of the data that were being generated on those molecules. Now importantly, these gains were made with a very small number of molecules being made or synthesized in the municipal chemistry wet lab. So over the course of all the series in the project, only 226 molecules were synthesized and from the chemical series that led to the development candidate, only 20 compounds were synthesized. And all that work led to the identification of our development candidate, SGR-2921. Now we believe this to be the most potent CDC7 inhibitor publicly described. This compound has been assayed in cell-derived xenograft models and patient-derived xenograft models and demonstrates really robust effects in AML models sufficiently so that we were compelled to submit this molecule for IND-enabling studies in preparation to initiate a Phase I trial in relapsed/refractory AML. So what I'd like to finish with this slide, which I think nicely captures how the computational platform was leveraged by the scientists to drive the program from inception to the identification of a development candidate. And I'd like to draw on an analogy of using a population of people greater than the world and being able to [ win ] that down to finding an individual. Starting on the left-hand side, the project team scored 79 billion molecules with machine learning. This is obviously a population greater than the world. But from that was able to select just over 24,000 of these for scoring biophysics based methods. This is similar to honing in on a particular town. And from that, making only 226 molecules in the medicinal chemistry wet lab, similar to honing in on a particular street in that town and then ultimately coming up with the individual we were interested in, the development candidate in a span of about 25 months. So at this point, I'd like to turn it over to Karen for the W1 inhibitor story.

So I'll wrap up on our last 2 case studies. So you've just heard from Hamish about the CDC7 program. As you know, several of our advanced programs are in oncology. We're also working on W1. We won't go into the mechanism of went a target in the DNA damage repair pathway. I think Hamish introduced you to that general space. It includes PARPs and other agents. Interestingly, W1 has been shown in the clinic to have important evidence of efficacy in humans. In particular, I'll just point out in [ emulsion ] carcinoma and in ovarian cancer, W1 monotherapy has been shown to be associated with 30% to 50% ORR. And those are 2 high medical need cancers. But in addition to that, there's evidence that W1 has potential in a number of other solid tumors. So the design goal here is really informed by, as was the CDC program, the activities in the landscape -- there are several companies, as you're well aware, going after W1. It became pretty apparent to us actually from the very beginning that discovering highly selective molecules for W1 that also have balanced drug-like properties, and I'll go into what I mean by that was important. But in addition, avoiding drug-drug interactions, we believe W1 as our many cancer agents will be important as a monotherapy in some cases, but will also be used to treat patients in combination therapy. So sort of starting with the end here in the sense that we identified, as we have set out to do, very selective, W1 inhibitors. I think you're familiar with AZD1775, a compound that had reached Phase II studies and also ZNC3, which I believe is in Phase I/II studies now. These molecules are all potent. But as you can see from these Kinomaps, the selectivity of these 2 prior molecules had much room for improvement, and that was one of our large areas of focus. The other that we felt was important, as I just described, was the drug-drug interaction liabilities. And it's pretty clear, I think, from the published data that perhaps similarities between the existing chemical matter had left this challenge unsolved with regard to [ CYP3A4 ] and perhaps other characteristics of the molecules. I showed you the endpoint of some of our chemical series, but I do want to go back and just describe what really, I think, exemplifies the power of running these programs internally at Schrodinger. After much work by the project team over the first couple of months of the project, we identified a very diverse set of compounds that were very unique actually. They look nothing like any of the existing W1 inhibitors, they were predicted to be potent inhibitors of W1 though. When we set these compounds out for the kinome scans, we were pretty shocked when it came back, what looked to be compounds that were less selective in fact than the innovator molecules, let's call them that for the moment. But coming together, the drug discovery team, the platform development team, it appeared to us that there were some patterns. And so we used protein FEP, which is where you are essentially perturbing the protein, not the compound in this case. So on the lower right, the white is the compound and above that, you can see side chains in the binding site changing, and those changes represent the characteristics of different kinases. It became very clear suddenly that we are able to use protein FEP to characterize different kinases and how they interact with compounds. We used that retrospectively and actually indeed validated that we could predict the profile, albeit dirty profiles of some of those early compounds. So we immediately leverage that finding to prospectively now interrogate large numbers of compounds, and you can see here, 445 million, not quite the 1 billion yet, but 445 million compounds that were enumerated and scored using this approach, FEP calculations on a fraction of those and 42 compound synthesized, 22 of which had what we originally were aiming for, which was exquisitely selective profiles. So again, just to recap, these highly selective compounds were predicted based on this protein FEP in a prospective fashion. We used a panel of 20 kinases to sort of check our work as it were. And in one cycle of synthesis, you can see that we went from these very sort of promiscuous compounds to extremely clean compounds and across a number of series. But as I've said, in addition to potency and selectivity, there are a number of other characteristics that need to be optimized, including oral availability, planability included as well as other characteristics. We're very happy to report that a number of our advanced molecules not only have solved this potency and selectivity challenge. They're also not [ T34-time ]-dependent inhibitors. So we took these molecules. Another big question was, well, we've come up with completely W1 inhibitors that they actually do what all the other W1 inhibitors were able to do. And indeed, we reticulated in a model that's been used for clinical compounds, the A427 lung model. You can see here that AZD1775, our compound were dosed for 26 days. And when you stop dosing, you could see that, yes, the AZD1775 adversative was effective. But once you stop dosing, the tumors grew back. In the case of one of our molecules, STC8188123, which was published at AACR, after dosing, the compound continued essentially to not just inhibit the growth but actually essentially cure the mice. Now the question, of course, going forward is, how do these work in the clinic. So we've seen this very durable single-agent activity in our preclinical models. And our goal now is to select the best-in-class compound with balanced properties and that preclinical characterization is in progress. And we are in the process of selecting our DC across these multiple candidates in this program.  Turning now to our last case study before I turn this over, the MALT1 program. So MALT1 is genetically validated, actually in a number of different immune disorders in people. But early on in the scientific exploration of MALT1, it became pretty clear that it's important in the BTK pathway or the NF-kappa pathway in B cells. You're all familiar with BTK, there's a number of agents on the market and many in development, addressing the question of relapse and resistance in B-cell malignancies. MALT1 is actually downstream of BTK and has a very important role in regulating NF-Kappa B signaling. This CBM complex is an important complex, of course, in not just B cells, but it also turns out in T cells and a variety of other cell types. Our focus for the moment has been on B-cell malignancies, but there is an emerging and important literature set about the role of MALT1 in Tregs and T cells in general as well as NK cells. And in addition to B-cell malignancies, there are potential indications in solid tumors in combination with checkpoint inhibitors and also in autoimmune disease. So I'm just going to show you quickly this is a complex protein. It actually operates as a dimer. If you can see that as sort of a mirror reflection. It has multiple different domains. And I've sort of used this schematic here to describe the design challenge. So as I said, it's a multi-domain protein, the paracaspase protease active site actually was identified quite some time ago by academics. The first compounds that actually were shown to inhibit MALT1 activity were peptidic in nature. This is a really large pocket actually and unfortunately, none of those molecules have entered the clinic. They are not, as you can imagine, very drug-like and are short-lived in blood. They were used though to demonstrate in vivo in animals that MALT1 has an important role in regulation of B cells. Sort of fast forwarding now to the second generation, if you want to call them that, small molecule MALT1 inhibitors. These are mostly allosteric, -- they're acting on the IGI domain, and that was the focus of our program. The initial allosteric compounds have entered the clinic, the first examples of which were not very potent and not optimized. Now, you've seen something similar to this in previous slides. Using our technology, we identified a number of unstable water molecules in the binding site. This is actually a very old molecule that was shown, I believe, in a phenotypic screen to find MALT1. This is, I guess, an antipsychotic that binds weekly to the MALT1 allosteric site. But you can see very clearly that there are a number of sub-pockets here where this compound doesn't reach, and this gave us some insights into how to drive additional potency into MALT1 compounds. So fast forwarding here to the progression time line of the program. We were able to identify compounds over the course once we have established the computational assay, we were able to identify compounds within a couple of months that not just had great potency that actually fell short a little bit of that target product profile, but they were sufficiently potent and also had excellent ADME properties that allowed us to go in vivo in, I would say, a very short period of time to establish whether these compounds indeed inhibited MALT1, we turn that in biochemical assays, but did they have efficacy and did they have the PK/PD relationship we were looking for. And that was a rapid breakthrough in the project. We knew then that the series that we were pursuing had activity. The question is, could we make them more potent. Over the course of the following 10 months, using a variety of our technologies, we were able to enter lead up 8 months and then identify the compounds that eventually became the DC and covered the TPP by 10 months. And all of that across about 80 compounds that were synthesized. You can see here a sort of representation of the multiparameter optimization challenge that we were pursuing. If you look at these 3 graphs, these are sort of representations of the balancing of solubility and permeability, solubility versus potency and permeability versus potency. And what's pretty clear is that the vast majority of compound score didn't meet all of those criteria. In fact, the compounds we were looking for were in this lower right quadrant. And after modeling around 5,000 ideas, we were able to identify 43 designs that met all criteria. And so returning to this sort of Google Maps world analogy throughout the course of the MALT1 program, we pursued 8.2 billion compounds scored with machine learning. And then the subset that was scored by physics-based methods is about 12,000. And we since sized 78 to find our development candidate. We have begun to benchmark our compounds against molecules that are out there in the literature. We do know that we have what looks to be the most potent MALT1 inhibitor. But as I have expressed throughout this presentation and Ramy as well at the beginning, we actually wanted compounds that had very well-balanced properties that included clearance, permeability of reflux, solubility, bioavailability, characteristics that we believe will lead to a best-in-class profile. I do want to point out that the site that we were targeting was one that meant compounds had lower solubility than in some other compounds. We had a very interesting collaboration with our material science group that helped us improve that solubility through formulation. And what you can see here is in vivo activity in an ABC DLBCL model, where comparing our activity of 1505 to ibrutinib, you can see that ibrutinib clearly works in this model. MALT1 inhibitors show signs of regression and when you combine these 2 agents, you see really exciting activity. So there clearly is single agent activity with MALT1 inhibitors. Recent disclosures from the J&J program do indicate that there are exciting and interesting responses in CLL and NHL in combination with BTK. Our IND is open, as we've previously announced, and we are initiating our Phase I dose escalation in relapsed/refractory B-cell malignancies this quarter. With that, I just want to wrap up by saying that we've given you just a few examples, vignettes of the types of programs we've worked on, where selectivity, optimization of properties for best-in-class programs, first-in-class, where we're going after modality switches, but where selectivity and optimization properties are also important. But I also want to sort of zoom out for a minute and look at the sort of history of our combined portfolio. You can see here that we've completed discovery and a large number in light blue of programs over the years. We have a large number of ongoing collaborations which we've already described. We gave you a couple of examples from Nimbus and Morphic, but over the last 5 years, we've increased the number of ongoing proprietary programs, some of which are partnered and some of which are wholly owned. We're very focused on improving the productivity of the drug discovery phase. And what we've looked at here is actually the metrics of programs over the last 10 years. We've compared that to industry standards as a paper that I think everyone references from Lilly scientists. And what you can see here is that the cumulative portfolio success which we anticipated would be impacted by this powerful platform is turning out to have a pretty profound effect on the success rate programs moving through stages, including through IND-enabling studies and into Phase I. Now of course, our data set is smaller than the industry, but we think this is a very promising trend, and we're excited to continue monitoring this as we continue to work on more programs and expand the portfolio. So just to reflect that, there are a number of programs in the clinic. I already told you about that. Not only have they moved into Phase I, several have now moved into Phase II. And in addition, our wholly owned portfolio where we've ideated around these projects, they continue to progress moving from the early stages to lead up and then over into IND-enabling studies in our first Phase I plan this year. With that, I'll turn the agenda back over to Robert.

Thanks, Karen. And really exciting to see the platform being applied in all these different ways. We're also very excited about how it can be further improved to provide even more value. The biggest area of -- or one of the biggest areas of investment is actually to increase the number and types of targets. Our platform can be used to progress. And there's 2 major areas of that investment. The first is to enable structure-based drug discovery for nearly all targets through combining some of these physics-based simulation methods with some of the recently reported de novo structure prediction methods, combined with basic structural biology investments. The other area related to this is to further improve our hit identification technologies, and we see avenues now to expand the application of those hit ID technologies to historically difficult and targets that have yet to be drug targets where small molecule binding is really unprecedented. The second major area of major investment is to further improve the effectiveness and efficiency with which our platform can progress drug discovery programs. And there's 3 major avenues by which we're aiming to achieve those improvements. The first is to continue to develop physics-based methods of the future, and that's related to investments we're making to more accurately model metals, where the quantum mechanics of metals can get pretty tricky due to the large number of electrons. Explicit modeling of more complicated quantum effects, things like [ Picato ] interactions, [ pipai ] interactions as well as further development of hybrid methods that build upon this integration of the machine learning methods and physics-based methods we described in some of the earlier parts of the presentation. Second major area of investment is to support a more comprehensive modeling of [ admitox ] optimization includes more explicit and more accurate modeling of things such as rate of clearance, time-dependent inhibition and [ hurt ] binding. And then the third major area of investment is further improving our de novo design technologies. So currently, we can support a valuation of -- on the order of 100 billion potential idea molecules, but this is still only scratching a small portion of the total idea space available to project teams. Chemical space, the amount of molecules a project team could consider synthesizing really is vast, and we're going to further invest in these technologies to more readily identify the best possible molecules for those teams to consider first synthesis and experimental characterization. I wanted to give a concrete example of where we're seeing these investments start to produce potentially very high-value future applications. Back in 2019, we published an exploratory use of protein residue mutation for energy calculations. So this is a technology very similar to what Karen described that was used to advance the W1 project. And we prospectively predicted that the L528W mutation might be problematic in clinic for BTK. That was an exploratory use of the technology. And it was very interesting to find earlier this year, we now have clinical characterization of that resistance mutation as being one of the more frequent resistance mutations for that system. And with this prospective success in hand, we're now taking a step back and asking for all of our projects where clinical resistance mutations might be relevant in the future, can we adapt this technology to try to produce molecules that might be more robust to the emergence of resistance could be a very exciting area for us in the future. Last major area of investment is to expand the applicability of our technology platform to new high-value areas, 3 are highlighted here. The first is expanding into preclinical development and formulation by investing in technologies that can help with solubility optimization, excipient selection and optimize things such as process chemistry that could help with small molecule scale up. We're also very interested in new modalities, including biopharmaceuticals, protein graders, molecular glues, very interesting technology development happening connected with all 3 of those areas. And lastly, we're, of course, continuing our big push into materials with a heavy focus on energy, chemical reactivity and polymers. Again, emphasizing the ways in which the technologies we're developing can be used in a number of different avenues. Back in 2017, we had published avenue to use protein residue mutation to improve antibody design, but a more recent capability that started to emerge as constant PH simulations. And we now think that by combining protein residue mutation with a constant PH simulation capability, we think we can start to facilitate design and optimization of pH sensing recycling antibodies, which could have all sorts of positive applications in reducing, for example, dose or increasing the effectiveness of antibody-based drug therapies. So very excited about the future. And with that, we'll turn it over to Jeff to discuss business outlook and opportunities.

Thank you, Robert, and it's an honor to be here with the Schrödinger team today. It's 50 days today since I joined the company. I guess what? It doesn't qualify me as a chemist. I won't be talking about chemistry. But I would like to share with you how we see the business from inside the company, which I think is quite different to where the investment community sees the business. I'm going to focus on how we create value from this remarkable platform that Ramy, Robert, Hamish and Karen have described today. So as Ramy started by explaining, we have 3 different ways of creating value. First is through software licensing; second is through collaboration and the third is by investing in our proprietary pipeline. These are quite different in terms of the contractual relationships we have ultimately with our customers. Software licensing really is arm's length relationship. Customers license the technology, then they deploy it as they see fit. We don't decide how, what, where they do with the technology. We have limited ongoing relationship with those customers. We do provide services you would expect. But we don't participate in any of the value created by the deployment of the technology through these sort of contracts. It does provide us with a lot of validation of our technology, and we have a lot of feedback from those customers. But again, we don't see any long-term upside in terms of share of products. Collaborations are much more -- a closer relationship with our customers. This is typically where some of the industry typically a big pharma or biotech company comes to us with a particularly challenging target or a difficult drug design problem, and they ask us to deploy our computational chemists, our internal scientists against that problem to help them fulfill their project objectives. Now in return for that, we get paid for providing that service, but ultimately, we also get participation downstream in that value. We have opportunity for milestones, future royalties, but the IP is fundamentally owned by the partner. Lastly, of course, we have a proprietary pipeline. These projects are conceived and driven by Schrodinger. We own the IP. We start the project. We complete all the internal computation discovery and development. It's an at-risk investment by us in building our own portfolio. Now of course, ultimately, we can out-license these assets as hits, leads, development candidates or even clinical programs or in the future, we can potentially retain some or all commercial rights. Obviously, we'll make those decisions on a case-by-case basis. And again, we expect to receive upfront payments, discovery development, commercial milestones and ultimately even commercial participation. Now, where we are today, as you've heard as mentioned before, we have roughly 1,600 customers worldwide. This is pretty much the entire universe of life sciences customers. We have 12 active collaboration projects right now, and we have 18 active proprietary programs. This includes those that are partnered already and those that we are retaining and advancing on our own account. Now let me just give you a sense of the business characteristics of these sort of ways of creating value. Software licensing is really a here-and-now opportunity for us. The revenue is already substantial. It's growing nicely. It's operationally profitable. It provides that validation. I mentioned it's a very extensive customer universe, and we expect continued growth via increased adoption. Collaborations, our near and medium-term opportunity, the work is funded by the partner. We see inflecting revenue growth from collaborations and proprietary programs. Those collaborations develop and enhance the platform. It's a limited and curated list. We're pretty careful about the number of collaborations we enter. Obviously, we have obligations under those collaborations that we need to fulfill. We can't instantly scale up our organization. So we take these very seriously. And over time, they cycle out. So we complete our obligations to the collaboration customer, and then we have room to add new customers. We see accelerating growth in the collaborations over time, but also in the number of targets in collaborations and frankly, in the value created as the programs advance, the milestones increase. So we expect to be more value there. And then from the proprietary programs, it's clearly a medium- to long-term investment. These programs are in investment mode. We expect to start to harvest them in the next couple of years. Working on these proprietary programs ourselves enables us to leverage but also enhance the platform. This is very targeted and carefully thought-out. We put a lot of thought into which programs we actually invest our resources on, and we do intend, of course, to advance into monetization, whether that is early development in the clinic or even commercially. So let me just share with you some of the metrics for these businesses, and this is now grouped in terms of how we report revenue. Our software licensing revenue for the first half of the year was $63 million. That was 25% above the comparable period in 2021. Our total active customers, as I mentioned, 1,600, but the 70% of the revenue comes from the top 100 customers, which, of course, as you would expect, is where all the big pharma companies are represented and our customer retention rate is extraordinarily high. And there are really no differences in these numbers having looked at them. Once people start using our software, it's highly valued by them, and we see very little attrition there. In collaborations and proprietary programs, we reported revenue for the first half of the year of $24 million. That was more than double the revenue for the first half of 2021. We can see that the number of active proprietary programs is growing steadily. Karen has already alluded to this number, including both partnered and unpartnered programs, we're now at 18. But what's particularly striking is the total available milestone opportunity, if you assume all of those programs are successful partnered programs that is that are already partnered is about $5 billion, and the number of programs now in development, eligible for royalties is 15. And clearly, there's been a massive increase in the number of programs that were both milestone and royalty eligible since 2018, and we expect that to continue increasing steadily. Just to summarize, I think all of the metrics that we track for our 3 sources of value are progressing very strongly. We are measured in how we deploy our capital. So we're not going to expand massively expand the number of collaborations in the proprietary programs, but we do think that we're creating an enormous amount of value for shareholders that we can expect to harvest over the next few years. So I think now I'll turn it over to Ramy.

I've just got one slide just to sort of summarize the day. I hope you'll agree that we showed you that we are the leader in molecular design software and have a real true track record of scientific innovation. And as Robert showed also a clear path to continuing to be the leader and continue to innovate in this field. I hope we also have done what I said at the beginning that you should be demanding from us that we demonstrate that the platform really works. I think we've shown some really nice cases of success. And I think that plot that shows or the chart that shows the pretty dramatic improvement on success probability is -- that should be something that we should be pretty excited about. That's a really important, obviously, feature of what the platform is accomplishing. You also saw that we have a strong financial position. And the balanced model, what we mean by that is, of course, this balanced model that Jeff showed the software licensing, the collaborations and the proprietary programs is really providing many opportunities for value creation in the future. So you saw, again, in the data that we have steady revenue growth from the software business. You heard about the announcement today of the new collaboration with -- or new partnership with Lilly, that's adding to our growing pipeline of programs. Again, Jeff gave you some nice details on that. And you saw, I think, pretty clear evidence to suggest that the revenue from the collaborations and proprietary programs are inflecting. So I hope you'll agree that we accomplished the goal of talking about the platform and the validation. And I really just wanted to close by acknowledging the really extraordinary work of our employees who have made all of all the things that we showed you today possible. So with that, I think we'll open it up to questions, Q&A. I think we're going to start here with any questions that are from people who are here. And then I'll remember to let Jaren allow people from the webcast also present. And then also I'm supposed to remember to tell everybody don't -- I think because people on the webcast won't be able to hear you wait until you get handed a microphone right and ask the question into a microphone, so everybody can hear. Oh, and everybody is here to help answer the questions. So I think I saw already some...

Thank you very much for the presentation. When you look at the software offerings and how it is utilized across the licensing side of the business, the collaboration side of the business and the proprietary part of the business. How does the software capability differ from business to business. And when you have someone like Lilly, who enters into this partnership, they've previously been on the licensing side, how does that relationship get managed? Yes.

That's a really great question. There are 3 aspects of the sort of differences, if you will, between what our software customers have access to and what we have access to in our proprietary programs and of course, in our partnerships. So one is that the scale. We've talked a lot about this. We have actually even -- this is something we can measure. The scale with which we are able to deploy our platform is on the order of 10 to 100x higher than what even our largest customers can deploy. What does that mean? That means those world map, the maps we were showing where we're showing we can analyze hundreds of billions of molecules score tens of thousands of molecules. That's something that our customers are just starting to be able to do. And you can see from what I showed at the very beginning, it's extremely important to be able to explore enough of chemical space to solve that multiparameter optimization problem. The other is -- this won't surprise you, I don't think, is that it can take a year, 2 years, even longer to productize new solutions. You saw from what Robert showed, we're constantly advancing the platform. That's technology that our group has access to, of course, before our customers do. And the final thing is -- and this is something we're trying to address, but this is still the case, is that, of course, the expertise that we have internally is very difficult to replicate at a customer site. We have side-by-side, next to the people who are actually deploying the technology, the whole entire development team that actually developed the technology. So that obviously presents a real opportunity to sort of -- to have that expertise that would be, again, difficult to replicate, but also we can solve problems in real time. That's what leads actually to the advances that Robert was showing is issues that come up during the program again can be solved in real time. And then again, going back to the second thing, that doesn't get put into the product right away. So anything else, Robert, you think that...

I think in terms of the -- I think you asked how these partnerships or relationships are managed. We obviously have a very long history of creating joint collaborative teams as well as the scientific collaboration we have, as you've seen, very many collaborations. And so we have an alliance management team. We typically have a joint steering or joint research committee with our partners to make important strategic decisions about the progress of the program, if that's what you were asking.

Yes. Did we answer your question? Now I'm worried that I didn't answer the question you were asking. So yes, get the microphone...

Yes. Just to extend off that. Lilly and their non-partnered parts of the business with you where they're doing licensing still. What are their capabilities on those projects versus the capabilities that are within the partnership?

Well, that's a great question and I think there are going to be 2 sort of benefits to the partnership. One is obvious, which is we're going to be developing medicines that actually help patients. But the other, and this is definitely one of the goals of the project, is you can imagine this kind of collaboration or partnership is an excellent way to transfer the know-how to Lilly. That's one of the goals. They have access to the software at a pretty decent scale, and they're going to have a front row seat, right, to our deployment of the technology at this huge scale and all this expertise that we just talked about. That knowledge transfer will be a part of the partnership. So it isn't there now. But we expect by the end of the collaboration or through the collaboration that will make a lot of progress toward getting them to the point where they can do what we're doing on the partnered program on all of their programs.

Gary Nachman at BMO. So thank you for the great presentation. So first, you highlighted the balance of potency, solubility and permeability that are some of the difficult components, what are the more challenging components going forward from here that maybe you didn't completely touch on as we think of the progress going more through clinical development and also into potential scale up manufacturing. So what are you thinking there? And then the new partnership for Lilly, for small molecule development, is that indicative of future partnerships that you're thinking of from here? And what's the actual capacity that you have for these additional collaborations?

Yes. So let me let Robert answer that.

Yes. So we're generally investing to predict all properties relevant to drug discovery. Some are clearly harder than others. Rate of clearance, for example, is an area where we're making big investments and having more predictive, more accurate methods to model. Rate of clearance will facilitate more rapid identification of those molecules that are best to advance. Likewise, you had mentioned preclinical development, also working to develop technologies that can support and assist with various challenges that come up in preclinical development. So broadly investing to support all of those activities, rate of clearance, if you want me to pick like a single property that I think is particularly tough. I might lead with that one, but that doesn't mean we can't make progress there. It just means we have to invest to make the progress there we would like to see. Karen, do you want to address this?

Yes. You also asked about the Lilly collaboration and whether this is indicative of future collaborations and what our capacity is. We described today that we're running about 30 programs between the collaborations and the proprietary and partner collaborations. And that number has been pretty steady actually. It's been growing, obviously, since the very beginning, but over the last few years, 30 programs is about the capacity for the discovery team. We expect because these programs last about 2 or 3 years, as those programs complete, we have the opportunity to obviously initiate wholly owned programs or to partner with external companies. I will say that, obviously, our business relationships have been evolving in terms of both the style in which we collaborate with people but also how much we take on. We want to optimize that and have a balanced approach to how we take on partner programs or also our wholly-owned programs. And the terms continue to increase. There's been a steady trend to increasing terms, and we expect and plan for that to continue. I don't know if that answers your question completely.

Yes... Sorry. Just remember to -- yes, maybe we should hold on to it.

Also specifically on the small molecule development versus other types of development programs, whether with antibodies or biologic type programs.

Well, I will say in our BMS collaboration, we did add a degrader component to that. We have a new modalities team that Robert's team is working on new modalities, including biologics. So I would say that you can expect to see, obviously, more small molecule programs and collaborations, partnerships, but I think you will increasingly over time, see additional modalities into that portfolio as well.