Schrodinger's Karen Akinsanya on NewYorkBIO's #VirtualBreakfast webinar series

Thank you very much for having me. I'm well, thank you, Derek.

Happy to do so. As you can probably tell from my accent, I'm from the UK. I was born in the UK, did most of my education there—secondary school, university, post-docs, and my first pharma job actually. I worked for Ferring Pharmaceuticals in the United Kingdom. I then moved with Ferring Pharmaceuticals to the West Coast of the USA, in San Diego, where I stayed for a while. PhD and postdoc in endocrine physiology and developmental genetics—quite diverse, and I think if you know me well, I've done a lot of different things that all lead up to what I'm doing now. After Ferring, I went to Merck, so I was at Merck for 12 and a half years here in New Jersey and in Pennsylvania. Finally, in my current role, I joined Schrodinger in New York City, which is a company that embraces technology, has created technology solutions that help us discover drugs. So I'm really delighted to be here and to share a bit more of my story and what we're doing.

Absolutely. When I joined Merck, I actually joined in a clinical pharmacology role. Although most of my training has been in basic science—I was an academic in the UK—one of the things I quickly realized at Ferring Pharmaceuticals was that when you are involved in the discovery of a drug, you don't often get to be the one that takes it into human trials. Fortunately for me at Merck, in my first role I was able to run human clinical trials, first-in-human studies. That was a fantastic experience. I think it actually made me a better drug discoverer because I really understood what happens when you take that fantastic test molecule that you co-invented and actually start to figure out what it does in people. That experience led me then back to the labs, back to the discovery labs. At the time, there was a very important initiative to embrace external science, which you're referring to—a lot of collaborations, a lot of interactions, at the time a lot with academics but then increasingly with biotech. I was fascinated by that, the idea that the lab was bigger than the walls of Merck and essentially by collaborating we could do so much more. So my diverse background in both basic research, drug discovery, and clinical pharmacology led me to a role in business development under Roger Palmer's team, really looking at all of the science that was going on externally and doing due diligence. It seemed like a natural fit for all of what I'd done. So I joined BD and learned a bit more about the business of science, which is a really fascinating space.

That's correct, yes.

Well, founded by a professor from Columbia University 30 years ago. In fact, this is Schrodinger's 30th anniversary. The notion that you could disrupt molecule design by using computers was a very long time ago. Fast forward to 2020, and I think that has not only come to life but is very active at Schrodinger, and actually at a lot of different pharma companies and biotechs using the software.

Great question. As we've already discussed, my career until that point had spanned discovery, business development, and clinical development. At the end of my tenure at Merck, I had the opportunity to go to a lot of different departments, which was a lot of fun. But when I reflected, what I really enjoyed doing was identifying those targets and figuring out how to test these ideas, how to test these therapeutic hypotheses in people. The idea that we could do that a little bit more efficiently, test more of these ideas, because as you know the failure rate is very high in drug development. The more ideas we can test, the faster we'll find those that actually translate into real therapeutic solutions for people. So when I came across Schrodinger, I was at Merck and we were actually using the software to solve a particular challenge in a program. It struck me that while one could use it on that specific idea, what if you could do this more broadly on a larger set of targets? Ramy Farid, the CEO, and I started talking, and the next thing you know I'm at Schrodinger running an 80-person group doing drug discovery, but also trying to figure out the breadth of the target universe that we can apply this software to. That's how I ended up at Schrodinger—this desire to do more with the tools that we have.

Maybe I'll just start by saying that drug discovery, drug design, has continued to evolve all the way back from the 1800s. It's been industrialized to the point now where we have very large teams of chemists and biologists working together to design molecules. The issue is, again going back to why we fail often and can we create improved molecules that are well positioned to test the hypothesis. Drug design is very complex. You've got potency, selectivity, the question of drug-like properties. The Schrodinger software is basically physics-based, it's atomistic level modeling of how molecules interact with each other. So imagine a universe in which you can explore vast amounts of chemical space on the computer, not synthesizing the molecules in an iterative empirical way, which is mostly how we do it, but you can ideate on the computer which molecules are going to bind with potency and affinity to the protein of interest. Then if you can do that on the computer, that gives you enormous headroom to design in all the other properties you need. Having the resolution of crystal structures or cryo-EM structures allows you to understand structure function and design these molecules with precision. The physics-based software essentially allows you to do that in silico. We use it to do druggability assessment right up front and figure out whether we're even going to be able to drug a target. You can imagine the savings there. Then refining all the properties you need into these molecules can be done using these physics-based methods. The result is that you can do this very quickly. You can identify molecules within months, then refine them and do multi-parameter optimization to dial in all the properties you need for the development candidate.

That's correct. I think the popularity of machine learning and AI in some domains is very appropriate—if you're doing image analysis, for example, there is a real advance. But when it comes to designing a molecule and figuring out how that molecule interacts with a protein, these are physics-based interactions. While machine learning can help in looking through existing data sets, it doesn't really help you when you're coming to the next protein that hasn't ever been drugged before or you don't have a chemical library for. When you go to these first principles methods, you can really make advances in understanding at the very fundamental level what these interactions are about. As you said, our teams are made up of chemists, modelers, and biologists. We have a multi-disciplinary team that understands what these calculations are telling us from a pharmacology point of view, because ultimately when you've selected these molecules with the help of the computer, you do need to test them. So that does require a cross-functional team. We do apply AI and ML to some of what we do, it's just not the foundation of what we do.

That's exactly right. I think interpolation is something machine learning is good at. But when you're in new territory, these physics-based methods are very powerful and allow you to move into chemical space that you simply—I mean, we're often asked how large is your compound library? All the chemicals in the universe is thought to be about 10 to the 60. We don't claim to go all the way there because a certain portion is not drug-like, but about 10 to the 40 is probably the size of chemical space. We're routinely running billion compound, multi-billion compound screens, which is obviously something you couldn't capture in a fixed library of compounds.

It's actually a fantastic catch. Obviously I came from Merck, which is an incredibly science-driven company, and my next home had to be a place where we start every conversation with science. As you pointed out, Ramy is a chemist, he was a chemistry professor. He actually was part of the team that created some of the solutions that we now share with the industry. The head of science, Robert Abel, actually discovered WaterMap and is very involved in drug discovery. I think our head of HR has a PhD as well. We have an incredibly talented team of 200 or more PhDs who are working on the software as well as this drug discovery team that's embedded. It's a classic pharma-trained, biotech-trained team, and our application scientists, the people inventing the software, it's a really highly technical group. It's funny for me coming from a pharma environment to a company where on one floor you have coders who come from Google and other companies who want to do something interesting with their code that can save lives and change lives. On one floor we've got this very tech-heavy group inventing the software, testing it, and making it ready for others to use. You have the drug discovery team, and also people who are helping others learn how to use the software. It's a really technical company. What's really important is that the goal is not to stop here. When you think about the challenges ahead in drug discovery, we only have structures for a portion of the human genome. As an industry, we've only drugged a very small percentage of the human proteome. There are about 20,000 proteins, and the drugs we have only address under 500 of those. We have a long way to go. One important aspect is how do we get better structures of all these proteins—there's a group working on that at Schrodinger. How do we do better crystal structure refinement? There's an ongoing effort to discover and develop new solutions for drug discovery, which extends to predictive models for other aspects of the drug discovery process. So it's a very science-heavy, research-heavy company that happens to have this software that allows us to also try and discover drugs either alone or in collaboration with others.

It's a very good question. Fundamentally, the software can be used to design molecules of all kinds. The history of the company is mostly around the life sciences and drug discovery. You're right to point out that if you pick up a magazine and you're looking at the structure of a protein, that was probably something someone used Schrodinger's software to create. So it's used very broadly. At least one portion of it is used very broadly for that, and academics use it in their research to figure out the beginning points of drug discovery. The question of focus is a good one because there is a materials science division, which is fascinating to me, something I'm not directly involved in, but there's a lot of potential in the design of molecules for materials science. The experience set, and this is something Schrodinger focused on about 10 years ago, is you're not really good at developing this software unless you're using it. So the focus at the time was to focus on the application of the software, not just hand it over to other people, but try to solve some big design challenges in drug discovery. That was done with the creation and co-creation of Nimbus, and after that came Morphic and Relay and a few other companies that really leveraged the software and allowed us to solve very big design challenges for some pretty challenging target classes. Through that learning, it became apparent that this has really important utility in drug discovery, so that is a big area of focus for the company. Looking forward, we are very keen to focus on areas of biology where there is at least some evidence of potential therapeutic benefit and to accelerate how we identify high-quality molecules for those. The question you raise about other areas and domains of application is something we're still working through. We have some really interesting and exciting days ahead in materials science and in other areas of drug discovery. Biologics is something we're really interested in expanding the use of the software for. There are a lot of biologic and peptide molecules that are very successful medicines, so that's an area we will be moving into. We have a very large research team constantly looking at these opportunities and figuring out if they're ready for prime time, ready for the drug discovery team or collaborators to work with. So there's lots of opportunity, and we continue to try to focus our efforts and deliver results on the pieces we choose to focus on.

Most people would take those numbers. Yes, thanks.

Yes.

I think we are a biotech company in some regards, but the software is a critical part of the company. If you were to ask Ramy this question, he would say we're both. We are still developing these software solutions, still pushing the envelope on what can be done with computational chemistry in many domains, as Jennifer just said. But we do have this established drug discovery unit, which at 80 people now is like a biotech. It's a piece that we think has tremendous potential but doesn't necessarily define the whole company. I think we really have to think about this as two strong pillars that will contribute to life science but increasingly contribute to other domains as well.

Yes, fantastic question. You have the history right there. Back to this concept that a lab is not necessarily defined by the walls of the lab, we are actually a completely virtual company. Our lab extends around the world, and that's something that during this pandemic has been pretty clear—when you have a global lab, you can actually generate data and keep your programs running. That's been really great. When you think about where science comes from and the ability for that science to impact patients' lives, it's impossible for any one institution to ring-fence that. So the idea of collaboration, working with others to pursue these therapeutic hypotheses and discover medicines, is in the DNA of Schrodinger, as you said. The philosophy still today is that when you have fantastic founders and innovators who've come across an important insight from a biology perspective and they want to go ahead and try to drug that insight, you can do that through a number of different routes. One way is to create new companies. Last year, a couple of new companies that are still in stealth mode were founded by Schrodinger with other important academic founding groups of scientists. We do believe there is still the need to collaborate with eminent and highly accomplished scientists who have these unique insights. We also think we have an incredible opportunity to help in walking through this proteome in identifying great targets internally and developing drugs ourselves. But we also think there are capabilities that pharma has that it makes sense for us to partner some of those programs for clinical development and commercialization. When you look at the collaborative framework for Schrodinger, you're going to see all of those things. You'll see newcos, several of the companies that were co-created have either done IPOs or are doing IPOs or are established companies that people know. Some are still in stealth mode, and we still have a lot of collaborations with pharma. Even though people buy the software, they want to use the work with the SWAT team, as we call it—the people who create the software and use it on a regular basis. So we have a lot of collaborations with pharma. You'll see all of those different modalities of collaboration continuing.

Absolutely, yes. In fact, the inbound interest in collaborating with Schrodinger has been quite active. Even since I joined, it's continued. Obviously this year Schrodinger is a little bit more on the map, and there is a large amount of inbound interest. You may have seen the collaboration with Twist, for example. As I said earlier, we're really interested in how one uses the software to design biologics. There's a platform or product called BioLuminate, but there are new research approaches the team is working on, and in collaboration with Twist we're pushing the frontiers forward. So absolutely, working with smaller companies and academic labs is still something we do.

That's correct, obviously very topical right now. You mentioned AstraZeneca, that's another collaboration that spans both small molecules and biologics. With Google, and actually a collection of pharma companies including Novartis, Gilead, and Takeda, Takeda basically initiated this multi-company approach to address antivirals for COVID-19. So those companies, Schrodinger, and Google Cloud have come together to bring each company's unique capabilities together to see if we can identify antivirals for COVID-19 for future use. Vaccines are going to be the first set of reagents we use, but ultimately an antiviral is going to be important. As you well know, the software does require quite a lot of compute power, which is one reason why it took a while for this to really come into its own. We now have access to clusters of GPUs, and we work closely with Google Cloud on that for our own programs and collaborative programs. Now excitingly, we're using it in this philanthropic alliance to go after antivirals for COVID-19.

That is correct. I think everyone sees it that way. Even if you have IV medicines to begin with for those who are most ill and in hospital, ultimately oral drugs are going to be important for ambulatory individuals with mild symptoms, severe symptoms, hospitalized patients at home. There are a lot of different ways that this can impact things.

Yes, it's a very good question. New York is the hub for lots of things, including patient foundations and an amazing array of academic centers. We have collaborations with alliances like the TRI-TDI, which brings together a number of academic institutions to find breakthrough ideas coming from those institutions where drug discovery is required. We work with individual institutions as well as these alliances. On the foundation side, the drug discovery piece at Schrodinger's internal ideation of programs is relatively new, but we try to focus on diseases where there are serious unmet medical needs and where patients are actually teaching us about the disease. While I can't point to a very specific foundation we work with on the disease side right now, we are very aware of foundations like Michael J. Fox—we're doing work in the neurodegeneration area and interact with other foundations that are doing an amazing job in identifying populations of patients with particular phenotypes who teach us that there are certain drivers of disease. Whether in New York or globally, these foundations seek to partner with drug discoverers to ensure we're focusing on the most important unmet needs and learning from the incredible work going on in genomics to understand the root causes of some of these diseases. These are areas we are very active in and hope to leverage to work on the most important diseases, sometimes for very small populations of people.

Okay, sure. First of all, let's take a moment to think about how we do drug discovery, and then I'll layer on the view of AI/ML and companies working in that space. As I mentioned earlier, we require a crystal structure that gives you atomic level insight into the protein, the pockets, and the active sites. When we have that high-quality crystal structure, the first thing we do is dock molecules—whether they be natural ligands, peptides, any kind of molecule that docks into a binding site. We can start these computational assays with these initial docking events. Once we find that we can reproduce wet data—if a paper says the affinity of a molecule for the protein is X, if we can reproduce that in this computational assay, we know the computational assay is accurate and the crystal structure is accurate. From there, we use that model to score compounds in silico. We're talking billions of compounds sometimes, scoring them for this particular active site or protein pocket. When you start a drug discovery program this way, you often find that the initial molecules you identify are either nanomolar or picomolar, meaning you have this incredibly potent hit. Most drug discovery programs start with a screen or an AI/ML model that finds the closest thing but not necessarily a perfect fit in the way these physics-based methods do. So the biggest difference is the ability to get accurate simulations of compound interactions, and the fact that you start with incredibly potent molecules means you're two steps ahead. Not only have you sped up that first part of the process, but we have about 25 programs we're working on, and with the ones run completely internally over the last couple of years, it's become very evident you can be in the later stage of discovery called lead optimization in under a year in some cases, even faster depending on the protein. That shaves off an enormous amount of time at the front end. Then going through lead optimization to get to a development candidate can be done in around two years. We have examples that were faster and some slightly slower, but on average about two years to get a molecule that is potent, selective, and has great drug-like properties. We think it's both faster and better. I've been in the industry 25 years, and being able to dial in all these properties in a couple of years for some pretty challenging targets is incredible. When you're able to design a small molecule for something that was originally drugged as a biologic, that's quite an accomplishment. Now, when we talk about AI and ML, they are very good at working with training sets. For example, if you feed an ML model lots of pictures of cats and a couple of pictures of another animal, it will identify that cat. But when you're working with a protein and you don't have the molecule in the training set that you need to initiate or finish your drug discovery program, it's very hard for that model to find that molecule in the dataset. If you're working on a related protein, the ML model will be able to find related molecules. We use it as part of our workflow because physics-based methods are computationally expensive. Using a combination of physics-based first and then ML to sort through the results works really well, but it starts with these accurate first principles methods. I hope that answers the question.

I think that's correct. There are certain parts of the drug discovery process we don't pretend we've disrupted. Safety assessment and all the things that have to happen up to an IND-opening study—we don't claim to adjust those. Having a more selective molecule and understanding the off-target proteins and dialing away obvious side effects is one thing, but at the end of the day it needs to be tested in vivo, and we don't claim to be able to predict everything. There are some amazing applications of ML and AI in pathology and text mining to look for patterns in the literature, but these first principles methods are really the foundation of drug discovery in identifying molecules and getting them into safety testing faster.

We have an active business development group led by Connie DeCruz. Inbound interest in collaborations is something we are very active in. The process involves outreach with a very clear therapeutic idea. If it happens to be in the life sciences—I'm sure there are lots of materials science companies in New York as well—but I can speak to the life science piece. If someone has a unique insight and they happen to have a crystal structure for a protein that they think will enable us to identify novel medicines, that outreach can come in through our business development group. Some of it is appropriate for the drug discovery team to look at, some isn't. People shouldn't share compounds with us; we don't want to see any compound structures, but protein structures are okay. It's really about talking with our scientists and our business development group about the opportunity to collaborate and figuring out if we can apply the technology. Your comment about focus is a very good one. We can only concentrate on drug discovery for a certain number of targets and with a certain number of partners, so we work through that process to see which opportunity is the right match.

Thanks for the question. One of the things I've always been passionate about is the public understanding of science. It became apparent to me when I was a PhD student and joined the Prince's Trust in the UK. I gave a lecture to a group of middle school students, and it became clear that there were children in that room who were clearly bright but did not see themselves as having a path to being a scientist, a doctor, or some other technical discipline. They were under 12 and already making statements like that, and I was stunned. I had a very privileged upbringing—I had a microscope and a typewriter when I was five, parents who could tell me what was going on down that microscope, both with training in healthcare and science. But it struck me that there are a large number of kids who probably have the inkling that this is an area they would love to explore and have careers in, but who unfortunately don't have access to what it takes. So with the Prince's Trust, I worked to mentor those kids. Throughout my career, I found opportunities to work with kids or with adults trying to make it in technology, science, and medicine fields. In New York, the National Medical Fellowships seeks to bring more diverse doctors into hospitals and healthcare settings, and I served on their board for a while. I'm very passionate about how we bring more science to kids at an early age, before they give up on the idea that they might be scientists. One thing I always wanted to do was teach kids science myself. My daughter was fortunate to do a gifted and talented program, and she was learning things that were years ahead of her class. That gave me the idea that what if we could do this for lots of kids? My Tech Learning was something I worked on with a few other folks—the idea of creating a lab where kids could come and try to explore science experiments, because a lot of kids don't get access to running science experiments on a regular basis. Giving kids more time with science, more time to explore and invent, is lacking in some way. So we tried to create a space for kids to do these experiments, and it was an enormous amount of fun. I'm not actively involved in it right now for reasons we're all experiencing, but one of the enlightening things was the number of letters I got from parents, thanking me for teaching their kids about immunity. Some of these kids were seven years old, and the fact that they understood what immunity was at that age, especially with everything going on now, their parents recognized how valuable that experience was. These days of COVID-19, with everyone talking about access to therapies, I think you're going to see a wave of kids who want to create vaccines because they realize just how important that is. So it's something I'm really passionate about. With all that's going on in terms of the social justice movement, getting equality in terms of access to great education is something I'm also passionate about. If you plant a tree and it has good access to light, water, and good quality air, that tree will grow and turn into a forest. Any one of those resources lacking, it's very hard for that to happen. So think about our kids that way—they need access to great resources, great education, great mentors. That's something I continue to be passionate about today.

Well, I want to shout out to Jenny Chambers, our head of education. She's great. She ran a molecular modeling course for kids recently. I don't know if it's still up, but the exposure kids get through her work—we work with a lot of foundations in New York that help educate kids in technology and software development. Extending that through to drug design with these online courses with Jenny is a really exciting development, given the lockdown, giving kids something different to learn while they're at home.

It is, yes. Schrodinger publishes hundreds of papers a year on the science, but one of the other things we're very active in is chemistry and molecular modeling and computational chemistry learning. We have a Schrodinger University. But this year, with the inability to hold live meetings where people come and learn, that's been shifted to this Summer of Science series, where we have a lot of our scientists, and I think some scientists from outside, coming in and talking about how to use the tools and about the challenges in drug discovery. So it's a fun series that's up and running right now, and I think it is open to everyone. If you go to the website, there's a way to sign up.

I think there may be some projects like that. Our head of technology, Pat Lorton, has been involved with some of the 3D goggles. There's something called the Looking Glass that allows you to walk through proteins and really look at how compounds are docking into proteins. We don't use that in the professional drug discovery group, but just that experience of running experiments and seeing what this is all about could be an amazing experience for kids. There are also large projects where large communities try to fold proteins and do other things. Not necessarily to discover drugs, but it's great for kids to get immersed in this.

If I had unlimited resources, I would say that just as we have coaches for running, soccer, and American football—let me get this right—I think just as we have coaches for all of those activities, I would love to see in every town and every city science coaches working with kids, showing them what science is all about at the bench, and talking about how we as a global community can improve global health by focusing on science and technology. Just making it available at a local level for kids all over the globe.

Thank you very much. Thank you so much, Derek and Jennifer. It's great spending time with you all.