S3, E10: Logica™-Drug Discovery Made Smarter with AI

Guido Lanza (00:02):
In the past, I would say the winners were the people that could design the best drugs. I would say in the future post-Logica world or whatever, post-Logica launch world, the people who can best define disease and measure disease at the efficacy, that's where the winners are going to be defined.

Gina Mullane (00:27):
The pharmaceutical industry is one of the largest in the world, generating trillions internationally each year, and yet one of this industry's most distinct features remains its high failure rate. The chances of success for a compound entering phase 1 trials has held steady at slightly under 10% for decades. In a traditional drug discovery framework, it can be difficult for drug developers to foresee challenges eventually encountered in preclinical and clinical trials, but what if we had the data and modeling technology to predict the journey of a candidate earlier in the process? Could we finally increase our success rate and get better molecules to patients faster? I'm Gina Mullane, and in this episode of Vital Science, I sat down with Julie Frearson, chief scientific officer at Charles River, and Guido Lanza, VP of integrated research at Valo Health, about the organizations' partnership, their groundbreaking Logica platform, and how the combination of Valo's AI capabilities and Charles River's data-generating power have inspired a re-imagining of the drug discovery process. Welcome to Vital Science, Julie and Guido. We are honored to have you. Tell us about yourself and your roles at Charles River and Valo.

Julie Frearson (01:49):
Hi, Gina. My name's Julie Frearson. I'm the chief scientific officer of Charles River. My background is in small molecule drug discovery, and I'm also intimately connected with the strategic partnerships, of which one is Valo Health.

Guido Lanza (02:11):
I'm Guido Lanza. I head up integrated research at Valo Health. What that means for us is really working on this closed-loop sort of AI-centric drug discovery platform, both internally facing as we apply it to our internal therapeutic programs at Valo, and then externally facing, in this Logica collaboration.

Gina Mullane (02:39):
Valo Health's use of AI-powered computation, paired with Charles River's drug discovery and development capabilities, is expected to transform the preclinical drug discovery process. Julie, how did this partnership come about?

Julie Frearson (02:55):
Thanks, Gina. Well, the partnership came about through several different paths. The first one, Charles River has a very, very deep relationship with a venture fund community, and the initial conversation was actually kicked off with David Berry, who is a flagship partner, and also happens to be the CEO of Valo Health. During that conversation, it became really clear that there was some real alignment in strategy between Valo Health and Charles River, and that we were both interested in providing the broader ecosystem with an industry best-in-class platform that would allow both effective and really efficient discovery of small molecules.

Valo Health had the AI-powered computation. They also distinguished themselves in the depth and breadth of that compute platform. They also had aspirations to democratize that platform for the broader industry, and that really spoke to our desire to do the same. Of course, we had the appropriate services, the data-generating power and the market channels, so together we actually made a really interesting pairing. Then, last but not least, both parties were open to entering into a very novel collaboration model, one that I don't think has really been seen in the industry to date, so the Valo Health-Charles River partnership was born.

Gina Mullane (04:45):
Guido, how will this partnership transform the preclinical drug discovery process?

Guido Lanza (04:52):
Thanks, Gina. I think, as Julie was saying, when we entered this partnership, this was on the heels of, for me, spending decades, plural, thinking about how AI and just data in general could fundamentally change the way we run drug discovery. As we were thinking through this in the early days, a lot of it was, we would have a CRO partner to take what the computer had done and put it into the real world, and it was very much arm's length. In a sense we had, I would say, DHL in the innermost loop, which clearly slows you down, but frankly, it changes the applicability. It means you're going to work on projects that the traditional process didn't deliver on, versus really rethinking what the process itself would be if you were able to offer this as a fully integrated solution.

The potential impact here, it goes beyond, "Well, I can get somewhere faster, cheaper, because now I have computers somewhere doing a virtual experiment where a real one would have been." It really inserts the concept of, I would say, intentionality in the different parts of the process. We do an experiment not necessarily just to advance wherever the drug discovery project is today; we might do an experiment to improve the ability to make predictions about the phenomenon. There is an intentionality, where, when you think about the traditional steps of drug discovery, early on you have this idea of screening to look for starting points, lead design to improve those starting, and then lead optimization to get to your drug.

Gina Mullane (06:38):
In my conversation with Guido, he explained how the partnership enables drug developers to infuse more data earlier, so they're not just screening out compounds that are unlikely to be successful, but also learning about how molecules may behave in their development path. Under this framework, the goal of early data generation is to develop an active learning loop to build predictive models for that compound and compound series. This allows drug developers to make more informed decisions about which molecules to pursue, and ultimately optimize for the clinic. In fact, the Logica platform has a 90% success rate in producing an advanceable lead series. It sounds like this partnership has the capacity to really transform the drug discovery process. Julie, how do you think it will benefit drug developers, and ultimately patients?

Julie Frearson (07:36):
Well, I think it's true to say these days that VC-backed biotech companies are actually responsible for drug discovery and development of more approved drugs per year than actually large pharma these days. There's been a flip in our industry, and the industry really relies upon the productivity and the innovation from these biotech companies, of which there are many thousands. A best-in-class compute-enabled drug discovery platform that is available to all of the industry, with palatable business terms, i.e., really accessible, I think provides an opportunity for these small companies to really efficiently develop and test their hypotheses, and, as Guido was saying, get the best molecules to patients faster.

We really are thinking about this from accessibility of the whole ecosystem surrounding us. Furthermore, it's increasingly clear that more and more drug targets are being identified from human clinical data insights. I think this is really the majority of where large pharma is looking for new target and disease associations these days, and I think a platform like Logica offers the opportunity to really develop molecules against a battery of novel targets, which we can then test in disease models, and ultimately in patients, to evaluate how useful they're going to be in a disease setting. One of the beauties of all of this is that all of the activities that I just described can happen under the Logica platform, and I think we're really empowering the biotechnology industry to deliver better-looking drugs in a more acceptable time frame, and that's got to benefit the patients at the end of the day.

Gina Mullane (09:51):
It does seem like AI is popping up everywhere more and more, and I'm curious; how is this integrated AI approach different from others that we may have been hearing about?

Guido Lanza (10:03):
I think it's a level of integration that essentially enables a scale that is otherwise inaccessible. What we're talking about here, if we think about most of the platforms that have begun to make a real impact out there in discovery, they're kind of working within the traditional silos. I might have something that lets me do some structure-based design better, or I might have something that very late on lets me compare two images of cells. Those are examples where I think folks have been quite successful. The scale and the level of integration here allows us to really take a step back and think about, what can I do to change the odds of success at any given point?

In a sense, the analogy that I like to maybe over-rely on is maybe like a gambling analogy, where essentially, when we sit down at the table or when we pick up a lottery scratch-off ticket, we know what the odds downstream are going to be. We know more, because we've built models of what will happen downstream in what would traditionally have been other silos, of success and failure, of the advanceability of those compounds, so it's not just about succeeding at getting a hit or at measuring a certain biology or finding a target; it's about the entirety of the process. It's about, in a sense, simulating what will happen downstream, not to the molecule itself in solution, but to the program itself.

What are the journeys? What are the things that are going to potentially derail us, and how can we make sure that we give ourselves the best chances of success, versus, I think, sitting down at a table where the deck is not stacked? That's, to me, the biggest piece. Now, in order to do that, you need a level of integration that allows you to start building those models from data that is generated later, traditionally, in the process. What I mean is, I need really good models of safety, and I want to use those models very early, not just at the end when I'm picking compound A, candidate A versus candidate B. I want them earlier, when I'm picking, which lead series am I going to go off and patent?

If I can simulate the likely causes of success and failure then, in this integrated way, and do that partnered with a data generation platform that is second to none, now I can continue to improve those models, and when they get ready, I can move them earlier and earlier in the process. That's something really I would argue nobody else can do. It's also the key to changing the economics and the expected cost, time, success rate, et cetera, that everybody talks about, that is necessary when we start to say, "Parkinson's is 80 diseases, not one; therefore, I have 80 drug projects, not one; therefore, I need a process that frankly has a different type of expected success in order go after them." It all links together, I would say, to what Julie was just talking about a minute ago.

Gina Mullane (13:16):
The example that Guido shared highlights that diseases which are often thought of in the outside world as a single disease may in fact be many diseases. This is why scale is critical to the Logica framework. By pairing Valo's AI capabilities and Charles River's data generation capabilities, drug developers can gain more insights faster. Logica can scan a broader universe of actionable chemistry, which includes billions of virtual molecules. Furthermore, Logica gives drug developers access to hundreds of in vitro and in vivo models and thousands of computational predictive models. Let's hear more from Julie on Logica's unique platform.

Julie Frearson (14:04):
I think this is a natural extension of what Guido and I were just referring to. We believe that there is no other platform available to the wide industry that is a match for Logica, the way it's constructed and its components parts, and I think the aspirations that we have for the platform are also pretty unique. I'd say that most AI platforms are either buried inside large pharma, or they're in technology companies. Frankly, you really do have to give up a pretty significant share of your downstream value as a drug developer to access those platforms, so there's a big, I think, accessibility problem in the industry.

I think there's also no other current example of a large data generator like Charles River and a compute innovator platform player like Valo Health coming together in this way, so I think a lot of the uniqueness comes from the combination of the power of bringing these two organizations together. We talk a lot about AI, and sometimes we forget about the human talent part of this equation. I think this platform, it also brings together the key human talent required to deliver success. Between Valo Health and Charles River, we have literally hundreds of years of combined drug discovery expertise. That expertise is critical in leveraging the predictive models that Guido was mentioning, but also critical in guiding programs to success.

Guido Lanza (15:48):
The idea is that there's sort of a so-what moment, which is, from the customer perspective, the fact we have, sure, the ability to model very complex biological signals or assays; what does that really turn into? If we think about the learn phase, it means we can leverage every experimental data generation capability that one might want to use in order to learn about the problem. Then, at a very, very large scale, once we've built these virtual models, we're able to look at vast spaces of chemistry, orders and orders of magnitude greater than what people would typically look at, for solutions likely to solve the ultimate discovery problem.

That's what I mean. When I mentioned earlier the intentionality of the data generation, that we can insert into the data generation, it means we don't necessarily need to try everything because we need as many starting points as possible. We need to run experiments so that we know we will perform really well in the following phase, so that our models we know will allow us to go through an optimization phase efficiently, so that we know we have interesting chemical matter ready to patent.

What we're really doing here is taking those models and then, instead of, say, maybe measuring a few examples of a compound and seeing which one's the most advanceable, or how they perform in some tier 1 ADME panel or something like that, what we can really do is look at millions of analogs of all of the series that we might want to push, against all of the most relevant models that we know are going to make the drug the drug. In a sense, we can say, for the series that are predicted to be the most potent against the biology that I care about, these are the ones that are likely to meet the least resistance as we push forward. These are the ones where there isn't a one-to-one correlation between the efficacy and the toxicity, or between the efficacy and brain penetration, or whatever the key attributes are going to be.

Gina Mullane (17:55):
As Guido mentioned to me, the goal is to really identify lead series with the highest probability of progression. In traditional drug development, this answer is not so easily found, or at least not so easily found in one place. The beauty of Logica is that it combines virtual chemistries with DNA encoded library and HDS approaches in the discovery of advanceable leads. It's an integrated approach that breaks down the silos typically found between data generating scientists and in silico scientists, so they are virtually working side by side. Julie, how does all of this happening behind the scenes translate to better drugs entering the market?

Julie Frearson (18:41):
Yeah. I mean, Guido's already clarified the fact that small molecule drug discovery is all about trying to optimize multiple parameters at once. It's one of the reasons it's so incredibly challenging to do small molecule discovery. You've got a relatively small organic molecule, with not that many features, frankly, and you have 12, 15 different parameters that you have to get right. As soon as you tweak one, often the others go off. That important concept that Guido mentioned about the opportunity very early on in the program to evaluate great expanses of chemistry design space is really the beginning of why you're going to end up with a better drug, because optionality in your chemistry is critical.

Often, when you're doing the early stages of a small molecule drug discovery program, and you have your first sets of interesting chemical matter, you don't ask yourself what's good about these; you ask yourself what's bad about them, and then you try and address that. We're going to be doing that on a much more systematic, systems-wide approach, and ultimately what that will result in is Logica's going to give us significant insight on those molecules that are going to make it all of the way, early on. This gives us the best window in getting safe molecules for clinical testing, in a relatively accelerated timeline. We're definitely going to be getting targets and concepts tested in patients much sooner than we otherwise would, I think, if we're not deploying a platform like Logica. We expect those molecules that are getting into safety studies to be inherently freer of liabilities and to have a smoother progress through that safety phase.

We shouldn't forget, also, that by taking this approach, not only are you getting better-looking molecules to the clinic, hopefully quicker; you're also recognizing early on if the investment in a program or a target is going to yield no results. I think there are plenty of reasons to suggest that molecules out of the Logica program will perform better from a safety perspective. Obviously it's the target biology that defines ultimately whether they're going to be efficacious against the disease, but it also has the opportunity to help us spend less time on failing programs, because I think the broad perspective that we're able to take, in terms of chemical space, will identify those programs that are simply not going to ever work from a small molecule perspective, so we'll spend less time on failing programs, which of course is better for the industry as a whole.

Gina Mullane (22:03):
I imagine that a lot of listeners will be interested in this more integrated system-wide approach, versus outsourcing specific discovery tasks. Can you tell us more about the solutions available to drug developers?

Guido Lanza (22:18):
Yeah. Again, in this end-to-end, but in a sense very simplified, kind of interaction that we're envisioning between Logica and the customer, there's really two points where we focus on these clear program value inflection points. I would say point A is, I have a lead series that is novel and that I can push, that looks advanceable, drug-like, et cetera, that will allow me to test the biological hypothesis that I set out to test. In a sense, Logical-AL, which is the first product in the chain, delivers that chemical matter. The second piece, though, is that it also delivers the models that are of sufficient quality to drive you through optimization in a much more efficient way. That takes us to the second piece here, which is the Logica-C product, where we're now taking those lead series and optimizing them to deliver a candidate.

There's a clear quality advantage, I think, that we've touched on, in terms of how much we've simulated the series, how pre-qualified they are based on their advanceability, which plays here, but clearly, if I can simulate the things that will go wrong, there's a significant time advantage that we're talking to also. The idea, of course, is, yes, we can get to these value inflections that are the standard; "I have a lead series, that's worth X. I have a candidate." We can get there significantly faster, with significantly higher probability, and frankly, maybe something that is the implicit so-what moment is this simplification of the relationship that we're looking for between the CRO and the customer. Really what we're talking about here is a simplified pricing structure. It is not based on effort; it is based on sharing in some portion of the success that we've added to, in terms of the portfolio value that we've added to the customer's portfolio.

Gina Mullane (24:37):
This transformation of the relationship between CRO and the client is not unlike the shift from volume-based care to value-based care in the healthcare industry. If a doctor helps generate better patient outcomes, they are rewarded for the value delivered, versus just the services rendered. Similarly, the Logica approach aligns all parties toward a single outcome, producing an advanceable lead series as efficiently and effectively as possible. Guido, if you'll indulge me, I'd like to revisit your gambling metaphor. Small molecule drug discovery is inherently a riskier game than large molecule, so how does Logica increase the odds of success in small molecule discovery?

Guido Lanza (25:23):
Yeah. I think maybe the model that I would think about here is, again, there is an expected investment that you make when you're entering the journey, the customer journey, going into a small molecule discovery. It is very different than what you enter when you go into, let's say, a large molecule discovery program, where the timelines, the error bars and the success rates are much more defined, and you know what you need to raise when or what you need to invest when in order to achieve certain goals. I would say that what we're trying to do is to make the small molecule experience mirror that much more closely. Frankly, the advantage over the large molecule experience is the success-based nature of what we're doing with the pricing around Logica-AL and Logica-C. In a sense, when I enter, when I start a small molecule discovery program, I know a lot more about what I will need to invest when, and frankly, it will be a much smaller fraction of the value I've already added to my portfolio than what typically happens, let's say, in a large molecule setting.

Julie Frearson (26:43):
I think I would concur with Guido that the business model that we've wrapped around Logica certainly provides any future partner with the confidence and understanding about what investments they're going to have to make to end up with a successful program. I would say there's also something to be said about the fact that this process is based upon a family of predictive models, some of which will be custom to the program, some of which will be generic to small molecule drug discovery, but having that suite of predictive models available to the program throughout its life I think certainly adds to the level of assuredness and certainty that a future Logica partner will have as they experience the platform, so I think there are two elements to that.

Gina Mullane (27:43):
What about the future? What do you see in store for this area of AI-powered drug discovery? How will Logica transform the drug discovery process in your mind, Julie?

Julie Frearson (27:55):
Yeah. I think there's a quality element to this. I mean, both Guido and I have spoken to the fact that we feel that we will be interrogating a lot more chemistry space much, much earlier in the process than you would through traditional models, and we've also talked about the anticipated time savings of executing on a platform like Logica. I also think Logica will genuinely change the balance between predictive science and wet science to enable a drug discovery program. More prediction in the mix. Much more focused make and test cycles. This will all add up to faster-paced programs. Now, the decision-making layered in on top still has to be exquisite, of course, but I think if you bring all of those things together, and I'm confident that the Valo Health and the Charles River team have all of the attributes needed to pull all of these things together and create faster-paced programs for our partners.

I think what we'll see on the ground will be a true partnership between computational and lab-based scientists. Decisions about next steps on programs will be an amalgam of the human experience and expertise and insight, from medicinal chemists who've been practicing their art for 20-plus years, layered on top of the machine's recommendations as to which of the identified molecules that we think have the best chance of success. We're going to see that working day in, day out, so I think overall we believe we have assembled the right AI tools together with the right human experience and experimental science, and this will create an improved path for small molecule drug discovery, which, as I said at the beginning, will be available to all comers. I think we can't over-emphasize the importance of sharing this with as much of the industry as we can possibly manage.

Gina Mullane (30:16):
Guido, from your perspective, how does the Logica technology pave the way for future innovation?

Guido Lanza (30:23):
Yeah. I think, in a sense, a lot of what we're doing with Logica is setting up the framework for how a data and AI-centric, but still heavily empirical and lab-partnered, process will work. Then, if we think about a lot of the areas of advancement in AI right now, things like AlphaFold, or what some people are doing with some next-generation Dell technologies, or with coming up with better targets through better translation and interpretation of the multiomic data and working backwards, et cetera, it all still comes down to the ability to solve the multi-objective problem that is small molecule drug design.

If I can use some amazing advanced AI-based structure-based design to come up with compounds, at the end I still need to be able to simulate their safety and their PK and their advanceability and all of those things. In a sense, what we're solving is a problem that gives you the infrastructure, frankly, off of which to hang a lot of these advancements. If I'm looking at some image-based phenotypic comparison of compound A versus compound B, in which one is having the effect that I want it to have, again, that is a piece of the puzzle.

A lot of the breakthroughs that are happening now, in those more siloed area, I would argue have the need of something like Logica to help translate those insights and really monetize those insights in the form of developing assets that turn out to be portfolio assets that at the end the FDA can look at, that we can advance and our partners can derive value from. That's probably the biggest impact we make is, when you change the economics of small molecule drug discovery, what else does it do? There are, yes, better-defined diseases. We're changing the economics of going after certain specific targets, but also, I think we're changing a lot of the framework for the way we apply AI here.

Julie Frearson (32:29):
The way I look at it is, Logica's about small molecule design. There's really no reason why we won't be doing a Logica for protein and antibody therapeutics of the future. The engineering space there is arguably much more complex in small molecules, and really the only way we're going to do that in a purposeful and intentioned way will be through AI and machine learning. I would also say that cell therapies are also a huge future application space, and I could imagine a Logica for cell therapy in the future, where we're using AI and machine learning to identify the best cells, the cell therapy applications, and you can imagine that will involve comparing deep, deep profiles of different cell types to be able to predict which will have best manufacturability, and which ones will be most efficacious and safe in the clinic. I think Logica is really just the beginning of our journey in how AI-powered drug discovery will apply itself across all modalities.

Gina Mullane (33:52):
Well, thank you both for this really exciting and informative conversation. I'm really inspired to see where this takes us in the journey of drug discovery. Thanks so much for your time today.

Julie Frearson (34:03):
Thanks, Gina.

Guido Lanza (34:05):
Thank you very much.

Gina Mullane (34:08):
Guido Lanza is the VP of integrated research at Valo Health, and Julie Frearson is the chief scientific officer at Charles River. Looking ahead to our next episode of Vital Science, in September, we'll sit down with Valerie Estess from Project ALS to discuss how several organizations came together to develop a novel antisense therapy to fight an ultra-rare and aggressive form of ALS. If you'd like to help support Vital Science, you can rate and review our show on Apple Podcasts, and if you haven't already, please subscribe to our show on Apple Podcasts, Spotify, or wherever you download podcasts. Do you have a suggestion, idea, or a great story to tell? Share it with us at [email protected]. Also, be sure to check out our sister podcast, Sounds of Science, focusing on innovation and trends in the life science industry. Thanks for listening to this episode of Vital Science. I'm Gina Mullane.