Why Deep Origin Is Betting on Both Physics and AI for Drug Discovery
The Harry Glorikian Show
Garegin Papoian, Co-Founder and CSO; Michael Antonov, Co-Founder and CEO
For February 27, 2024
Final Transcript
Harry Glorikian: Hello. Welcome to The Harry Glorikian Show, where we dive into the tech-driven future of healthcare.
Investors and companies in the life science industry have been betting a lot of money over the last few years on a single idea.
If you boil it down to its simplest terms, it’s the idea that computation will help us get a lot better at developing new drugs.
The argument is that we get more and more data about the DNA and RNA sequences of single cells and whole organisms, we can use software to figure out exactly what’s going wrong in various diseases, at a molecular level.
And then we can do some more computation to design or discover drug molecules that will interrupt those disease processes at just the right moment, and prevent or repair problems inside cells without any bad side effects.
But the word “computation” covers a pretty broad range of techniques.
And the reason that there are dozens if not hundreds of computational drug discovery startups popping up is that everyone has their own hypothesis about what specific kind of computation is going to be the most powerful.
For example, you might be convinced that the most important thing is to understand the physics of protein-protein interactions, at an atomic level.
And so you would put your money into atomic-scale simulations that show how proteins fold or unfold to form different shapes under different conditions.
Or you might think that it’s more important to model proteins at the molecular scale, to make predictions about whether and how a particular drug molecule might dock with a target protein.
Or you might think that it’s smarter to try to model whole cells and see how different molecular pathways interact to affect different functions of the cell.
Or you might not care about the details of physics- or chemistry-based models at all.
In that case could just take a big generative AI model, similar to a large language model, and train it on huge amounts of unlabeled data about genes and proteins in diseases cells and healthy cells to see what kinds of predictions it comes up with.
It’s too early to say which of these computational approaches—and which level or scale of focus—is going to be the most fruitful.
But maybe you don’t have to choose. Maybe you can bet on all of these different ideas, all at once.
My guests this week are the CEO and CSO of a startup that’s taking an all-of-the-above approach.
It’s called Deep Origin, and it was formed last year from the merger of two companies founded by theoretical chemist Garegin Papoian and software builder Michael Antonov, respectively.
Antonov helped to found the virtual reality hardware company Oculus. After Facebook acquired Oculus, he got curious about longevity and how software could help untangle the trillions of gene-protein interactions that mediate health and disease.
He founded a company called Formic Labs to dig into that problem, and last year the company changed its name to Deep Origin.
Papoian is a former academic scientist who’s who also took the helm as CEO of his startup AI and who’s interested in how to use software to model molecular dynamics and quantum chemistry.
Recently Antonov and Papoian decided to join forces, and Biosim AI merged into Deep Origin.
And they say the company’s philosophy is that physics-based modeling by itself won’t be enough to build a powerful drug discovery engine.
But neither will generative AI, which requires more training data than lab scientists will ever be able to provide.
So they think the only reasonable approach today is to combine the two, and use both physics and AI to try to get better at predicting which molecules could become effective drugs.
Exactly how Antonov and Papoian came to their conclusion, and how that integration is playing out, was the main theme of our conversation.
It’s important stuff, because if Deep Origin is right, then a lot of other more specialized biotech and techbio startups could be going down the wrong path.
So let’s dive in.
Harry Glorikian: Michael, Garegin, welcome to the show. It’s, uh, it’s great to have you both here. And I feel like it’s. I’ve been thinking about this, having you guys on the show for a long time, but I’m glad we finally got it scheduled and got you rguys on.
Garegin Papaoian: It’s a pleasure, Harry. Thank you for inviting me.
Michael Antonov: Happy to be here.
Harry Glorikian: So I want to, I guess, give everybody sort of a at least a level set on, on, on your backgrounds because, uh, Michael, like you did not start in drug discovery and biological sciences if, if, if my homework is correct. Um, so can you start like, a little bit about your background so people can understand that and how you ended up in this space.
Michael Antonov: Yes. So I actually originally was a software programmer, very passionate about computer games. So built two. Was a CTO or technical lead in one of a game technology companies, and ultimately was a part of the Oculus VR journey, where I did lots of infrastructure in the case for the Oculus Rift, computer vision and other technology stacks. So for many years I was really into building software platforms. And ultimately when I was in, um, actually rejoined Facebook after the acquisition, I was trying to imagine what is the most meaningful thing I could do with my life. And I got very, um, interested in longevity. And because the genomic revolution has happened with a lot of noise, and I was curious about just how much do we understand it. And the other part which became very interesting to me is how do our bodies work? So I ended up taking lots of courses. I remember biochemistry was like the hardest thing ever while working full time sitting with pre-med students. And that’s probably the type of thing which is also drove Garegin’s passion is being able to see at that point animations like of the ATP and how all this cellular machinery works, like in small microscopic world. It seems predictable, it seems something we can understand. And that got me very interested in simulations. But also what can we do for longevity? So long story short, I then started investing in the space, very cautiously. I realized it’s quite risky, um, and learned. And ultimately decided that we need a lot better scientific tools. So you imagine even like a single cell can have 10,000 expressed genes, there is trillions of them. All the interactions. There’s no way a human is going to be able to get it and understand it. And if we are going to truly improve health span, we need to be very kind of multimodal, attacking many pathways and things at once. And the only way you’re going to do that is if you can simulate or understand the system. So I’ve really dedicated and decided that a lot better software and infrastructure needs to be built to speed up this process.
Harry Glorikian: Well, the one key takeaway I’m taking away from this is it’s never too late to go back to school and start a whole new career in a whole new space. So for anybody that’s listening, you know, think about that. Garegin—I mean, you’re more. I want to say grounded in the space. But I know, like, you’re super interdisciplinary. So like a unique background of components that are coming together to attack what you’re what you’re trying to attack now. So can you go a little bit into your background a little bit?
Garegin Papaoian: Yeah, absolutely. So I’m a theoretical chemist and I’ve been most of my career and I’m still in academia part time or on a part time sabbatical. Theoretical chemistry is itself an interdisciplinary area that sort of is part chemistry, part physics. And in the last, uh, 40 years, it’s also become part computational science, where, uh, we have to also program, uh, for those of us that are quantum chemists, you program quantum chemistry codes. If you do molecular dynamics, then you program molecular dynamics codes. And just to give, uh, sort of a perspective on how complex these things have become, one of the leading molecular dynamics codes in the world is called GROMACS. And it has 2 million lines of C++ code. So it’s a humongous project with lots of moving parts. Probably no one fully understands how it works at this point. It’s just so big. And so it sort of requires not only just knowing equations of chemistry, equations of physics, but also how do you, what are the best software practices? How do you develop a complex, complex software product and maintain it and documentation? So I’ve been, and you are right, so I’ve been interested in not just in chemistry but uh, in physics, in computer science and uh, and over many years, uh, I’ve done some paper and pencil papers, um, that I enjoy, just analytical papers. But most of my work has been computational, in computational chemistry, computational biology. And, and I feel that my, I enjoy, and my strengths are being a tool maker in this space. So I like building complex new tools in computational chemistry and biology and then applying it to novel scientific problems. And a few years ago, um, I realized that, uh, that I over my career, I participated in many projects, but they also, they have this dissipation tendency that after you move on, then this software decays and maybe it’s not maintained well and so on. So, uh, so there was an idea that, uh, I bring my, uh, experience in developing new tools in computational chemistry to work on problems of drug discovery. So that’s how that’s how I launched Biosim. And then pretty early on, I met Michael and we started early on working closely together, having a shared vision and also helping with each other in many ways. And yeah, so that’s roughly how it started. And I feel that a startup company allows many new opportunities for this software development, and that can be applied both to health and longevity that in academia would have been hard to accomplish. So yeah, so it’s a new life for me too, actually, in a way.
Harry Glorikian: Yeah. So I mean, for for everybody that’s listening, I mean, I would, you know, I was I’ve been watching these companies for almost since inception. So uh, been watching how they’ve evolved and grown. But I think you guys have had to put a unique team together. Not just say, what I consider drug discovery and say AI but a real understanding of, as you said, Garegin, was artificial intelligence, molecular dynamics, physics, and a number of other disciplines to sort of bring something together that then can get you to the answers that you’re looking for. I mean, how do these, how do you look at these two fields intersecting in your approach to drug discovery, what is the advantage that this interdisciplinary expertise offers that say a normal approach may not.
Garegin Papaoian: Yeah, definitely. Great question. Um, so historically until this recent revolution in AI, the approach in computational modeling and simulation was physics based. And similar to for your listeners, uh, to, to make an analogy, similar to maybe weather modeling that, you know, like every day you open your iPhone and instead of says, oh, tomorrow it will rain, or maybe even in a week. And the question is, how can they predict that? Right? And it turns out there are these super complex simulations run on huge supercomputers that take the whole Earth and atmosphere and put it into small grids and just sort of model how air moves from place to place and, and heat transfers and from all that they can then locally, for you, tell you that you’ll have snow tomorrow. So it’s pretty remarkable that they can do that. But there are limits to, for example, you cannot predict weather for the next year, like there are limits to how far into the future you can predict. Even though you have the physics equations, but they sort of degrade in a sense. The predictions degrade if you go into longer time horizons. So there are limits to physics based modeling. There are similar limits to physics based modeling in biology and chemistry. There are some things that we can do really impressive job and get and get nailed answers. And you check it with experiment and you get to the fourth or sixth digit accuracy in terms of the result you predicted and the results that the experiment measured. But then typically, especially in biology, most things we cannot predict that well. Proteins are complex, DNA is complex. And then they all come together. And from this sort of mess that’s very dynamic, and it turns out that purely physics based modeling has really hard time to just go and photorealistically model these things. Maybe 100 years from now it could or quantum computers, I don’t know. At least for the next few decades, that’s out of the question. And, uh, and then the AI came along, the AI revolution ten years ago. And I’d say, I feel that there was maybe a bit of a, sort of, there was this huge shift that especially in the industry, but also in academia, but largely in the industry where, uh, lots of physics based modeling was forgotten quickly and everyone was just wanted to take data and just put AI on it and quickly get low hanging fruit. It was appealing to investors. It was appealing to like, we can just use the magical AI and get something done in a month that was unsolvable. So that was the test. In a sense, that’s the situation today to some degree. Uh, but I feel that, again, in biology, you depreciated that data. Like, every data point you get through sweat and tears and money. So it’s not easy to get like 10 million data points for your AI to have plenty of data. In some problems, you need more than 10 million. Maybe you need 10 billion data points and it’s completely out of the question. So so that’s where I think we are forced to sort of reckon that you cannot just do. In some problems you could can do, maybe there is enough data, maybe the problem is narrow enough. But in most problems in biology and biomedical research, you somehow have to combine the two. And that’s my philosophy and I believe that’s Michael’s philosophy.
Michael Antonov: Yes. That’s actually part of the reason we have these kind of two components to the company is because we realize whatever is done in simulation, it needs to have experimental validation as a part of it, but also [for] certain things you do need to have lots of software infrastructure to even track experiments. I think the state of the art in many areas today is that you do have databases, and you may have limbs and certain things to know a bit about your lab, but at a certain level, your lab, your software does not actually know or understand what is the data it captured. So somebody may put in a spreadsheet with like a results of plate readout for different, like what are the different things in each well for each cell type, but they don’t know what the columns mean. Therefore software cannot come in easily and interpret it. So lots of bioinformatic work needs to be done. So that aspect is not going to go away. Hopefully it’ll become more optimized as robotics simplified. So at the same time these two systems need to be married. So if we are going to fully close the loop on this integrated “simulations to wet lab to understanding” cycle, then we need software which looks at all of these parts and makes it accessible. And that’s really the way we look at it. I’m really excited about the simulations and really kind of AI and physics bridging the gaps for broader models, I hope. One thing I like to talk to Garegin about is this molecular self-assembly, which I define it a slightly bigger problem. So because you have a protein folding, you have interactions which may be MD handles, but can we actually throw a volume of molecules with DNA, RNA and everything in a soup and have it do what biology does? Could that be done as in, say, the next five years? And clearly that won’t be just physics. It needs to be physics and AI and things. But if we can really do that, then we can truly understand biology. So kind of marrying that and supported by experiments is very exciting.
Harry Glorikian: So, yeah, I mean, you know, I love this space because there’s not a day that goes by where I’m like, I didn’t I either I don’t know something and I gotta learn or I’m learning something new, which is what keeps the space exciting. But, you know, I’ve talked to lots of people on the show. I’ve talked to lots of people, you know, new companies and so forth. What makes the approach that you guys are taking to drug discovery, say, unique compared to and I’ll just say traditional methods, not to pull all the new stuff into play. What the twist? What’s the secret sauce?
Garegin Papaoian: Yeah, I think we because we have bought the simulation part and the life sciences platform, we probably want to have, uh, two, two answers to that question and then the totality of them coming together. On the simulation side, our philosophy is that we need to develop new tech and that that can advance the field. And in particular integration that, you know, if you think about a cell, you say, okay, a cell has a billion proteins, but then you zoom in and then you have maybe 2 or 3 proteins coming together in a complex configuration and then use the mean, and there’s a protein pocket and you start to see side chains and they are rotating and you start to see hydrogen bonds. Well it turns out that you at least in today’s world, you cannot have a single model that describes all that, although it’s a single reality. But just for like as again, the same thing as we do, you do local weather modeling. You do maybe for the country. You do it for the planet. Like these are sometimes these are different models that maybe talk to each other. It’s similar for cells too, is that we have a model that maybe with quantum chemistry models, fine details of how hydrogen bond is made between two small molecules. And then you go to the protein level. That’s another set of models. And then you go to the cell scale. It’s completely another set of models. And one of the. So most companies typically focus just in one of these spaces. And because of that they sort of miss like you might focus on the pocket but sort of not see, oh, there’s a big protein coming. And because of that your pocket change and, and the calculations you did sort of doesn’t make as much sense because you ignored that interaction and so on. So, so one thing that I think distinguishes us on the simulation side is that we have four teams working on four different scales, from small molecule docking to molecular dynamics of proteins at atomistic scale, then coarse grained modeling of multi-protein assemblies, and then whole-cell modeling. And these teams then can work with each other, make in projects to make sure that there is consistency and then bringing AI to all that. So I think that technology stack, the depth of it or the breadth of it, is something that I think is quite unique. And, and the combination of physics and AI is, is powerful too. So we think that we are developing a powerful new overall technology stack.
Harry Glorikian: Yeah. I was looking at some of the slides that you sent me and thank you, because that completely changed every question that I had that I originally came up with. So, uh, it, it brought me up to speed on where the company is, but. You had a couple of slides there where you’re sort of, you’re saying the AI enables much more faster and say accurate virtual screening. And so is that a function of the code? Is that a function of the tech stack you guys are using? Is that a function of how you’ve assembled the pieces to get you there, compared to some of the other systems that you had on there? Just curious of that technical advantage because it I assume it means less GPU time and therefore a lower cost.
Garegin Papaoian: Yeah, again in that case, too. Harry, it’s it’s, uh, it’s a combination. It’s not pure AI. We cannot actually get these results with pure AI. Meaning if we just come and say we don’t know physics or chemistry and we don’t want to know, we’ll just throw the data into, like, some super powerful AI and it will all figure it out for us. I don’t think we could get results like that with that approach. So it is heavy physics there too. But then we sort of know in important places where physics has gaps, we sort of plug those gaps with AI and, and then it becomes almost like an artisan work. You just have to know that, what are the strands of physics and but also what are the weaknesses and, and where AI can glue in. Because sometimes they cannot they might not be completely compatible with each other. So it’s a very artisan work of basically assembling the architecture, let’s say, of that virtual screening algorithms and software. And a part of that is also is the question of we have really strong teams in these areas, uh, very talented folks. And uh, and I think that’s part of the they just, they learn and uh, have background in sort of having that artisanship in building software like that. Um, yeah. So I think that’s the. But that that’s that would be my answer. There are infinitely many ways how you could combine AI and physics. Even in AI, as you know, there are many ways, like there are many AI approaches, some of them are better than others. And then there are also lots of data issues, like what data do you take? How do you clean it? Like lots of sort of small questions that maybe each of them is not a huge thing. But if you if you string it together, even sort of two similar teams working on a similar problem could end up with very different solutions just because they answered those questions differently along the way.
Harry Glorikian: So, you know, I’m going back to the first conversation we ever had. And I don’t think the premise has changed, but this whole concept of undruggable targets. How is it that the system that that you two have created or brought together enables better drug design to to tackle this challenge. And just to add on top of that, right, how do you identify high quality hits when you’re doing that? So say, you know, maybe you r could both tackle the answer to that combo question.
Garegin Papaoian: So I’ll start with the second one. Um, obviously the ultimate way to check whether do you have a hit is by experiment. So you could have an in vitro or an in vivo experiment and have an assay and test and see if your computational predictions come true. You could synthesize a molecule and then bring it to, let’s say, an SL assay or a binding assay and see if, if, if you see the outcome that you are expecting. Interestingly enough, to even their computation can help. At the end, you still need an experiment for sure. That’s the ultimate answer. But as an interim step, there are these computations called free energy perturbation in molecular dynamics that for example, in drug binding you have your protein, you have your putative drug in the pocket, and then you have water around it and ions and metabolites, anything you want. And then you could do this, uh, advanced calculation in molecular dynamics where you say, what’s the binding energy of that? And it turns out that, uh, that it’s not an easy calculation to do. And computationally, it’s expensive too. But, uh, but you might be able to get an answer to 1 to 1.5 kilocalorie accuracy. That. For example, for heat discovery stage, when you are looking at many virtual molecules, that could be good enough. And it might allow you to, when you select computationally molecules that way, when you go to the experiment, then you get really high hit rate in terms of the experiment. So so that’s uh, maybe I’ll, I’ll, uh, stop there. We can come back to the undruggable targets. Uh, let Michael chime in.
Michael Antonov: Well, actually, I know it’s a little bit of a side, uh, comment, but the thing which, uh, got me really excited working with Garegin and looking at the industry is building this multidisciplinary team where well, first of all, Garegin himself has had experience building both these low level models and high level, coarse grained models that he’s built this, uh, actin myosin formation simulation, which I had a chance to take a look at academically, which was quite involved, which is a high level abstraction. And we also have a team, uh, one person, Jonathan Carr, who’s actually made the first full cell model in the pathway simulation ten years ago. He was one of the original people in Formic Labs, uh, which is our, um, kind of the first part of the company on our side. So we have the idea was to bring these researchers with different levels of experience, of knowledge in these different domains and see how they can collaborate and integrate. And the idea is, as Garegin described, is that these different models and systems will support each other. And I think the other key aspects which makes this company unique is we are thinking about it as a platform. So with Garegin’s leadership, we have this really, truly unique research; nuggets of really good technology where we can be state of the art in certain parts of discovery; but at the same time, we’re thinking about it as a platform where we have general compute resources, general infrastructure where these algorithms can run and we can provide, whether it’s through Jupyter notebooks or simulations or easier tools, broader accessibility across and also integrate with these later stage or different experimental tools.
Michael Antonov: So all the kind of data sets and infrastructure is available across the board. Though the idea of a company then, is to provide this platform and framework where different scientists can collaborate within the same environment and in the future even provide their own plugins and tools. No matter how good we are at simulation, no matter how good we are at any part that we do, it’s a very broad industry, so we’ll want to integrate with other tools, but also provide the ability for people to have components or aspects of this which are, you know, which we did not create. So as a part of a vision of a company, we want to build this more open ecosystem around making some of these things accessible. Of course, some aspects of state of the art will be unique for drug discovery and supporting, you know, higher value clients. But we want to think about how much, how can we open this ecosystem and cover this range to make it accessible.
[musical interlude]
Harry Glorikian: Let’s pause the conversation for a minute to talk about one small but important thing you can do, to help keep the podcast going. And that’s leave a rating and a review for the show on your favorite podcast player.
On Apple Podcasts, all you have to do is open the app on your smartphone, search for The Harry Glorikian Show, and scroll down to the Ratings & Reviews section.
Tap the stars to rate the show, and then tap the link that says Write a Review to leave your comments.
On Spotify, the process is similar. Search for The Harry Glorikian Show, click on the three dots, then click “Rate Show.”
It’ll only take a minute, but you’ll be doing us a big favor, since positive ratings and reviews help other listeners discover the show.
Thanks. And now, back to the show.
[musical interlude]
Harry Glorikian: It’s probably easier if we had slides, but, you know, maybe you can walk us through any case study where you were able to show some superior performance, let’s say.
Garegin Papaoian: Yeah, I could give a quick example. So we were, uh, we wanted to test our virtual screening methodology, and there are multiple ways to test it. One way is purely structural. For example, you take a compound, you dock it into the pocket of a protein, and then you if you know the crystal structure, you see, well how well did I do? Like is it more or less identical to how the ligand was bound in the crystal structure, or is it in a completely wrong place? So that’s one way how you could identify quality of docking for example. Uh, but another way is actually screening, which is if I take, and that that’s one of the slides that I sent you, Harry. So so for example, there’s this protein called Jak2 and it’s biology. Let’s put it aside. But let’s say it’s a protein that’s important in disease and you want to target it. And uh previous works already identified about ten compounds that, uh, were active with this molecule. And the question is if we in virtual screening, we are going to take millions or sometimes billions of potential molecules and find sort of the needle in the haystack. So the question is, if I take, for example, ten known molecules and I mix in 100,000 random molecules. So then the dilution is 1 to 10,000. Your chance that you are going to even like for example, when you do, if you randomly simply just shuffle these things and take top ten, your chance of finding even a single active molecule is minuscule. It’s rextremely small. So we did a docking study like that and we found that out. So again we are mixing ten active molecules with 100,000. And we docked and we scored using this AI based scoring we developed and then ranked. And in the top ten ranked list, our ranked list, three were active. So that sort of shows a huge enrichment. I think when we calculate what is the enrichment, it’s about a factor of 300. So if you had to do 100,000 experiments this allows you to do 300 times less experiments, which is I’m sure many labs will take that. So this is one way, uh, that it shows how computational drug discovery can tremendously accelerate research in the lab by just, just narrowing down things you have to synthesize and test in the lab.
Harry Glorikian: So. But what are the, you’ve done that in one thing, but what are the challenges potentially in scaling up the technology for more customers, because I’m assuming you guys are taking on multiple partnerships, but, you know, you got to scale that technology up. And all of us know that scaling, doing it in once is one thing. Doing it at scale is is always a different challenge. So what are the challenges you guys are facing while you guys are trying to do that?
Garegin Papaoian: Maybe Michael can take that because scaling is the life sciences platforms domain, I’d say in our company.
Michael Antonov: Well, there’s different parts of this. There’s a first set of, uh, partnership customers, we do provide a meaningful amount of kind of hands on help, to make sure all the results and everything is good. At the same time, we envision a more open self service aspect of it. So for example, um, right now it’s seen as a we have two pathways. One could use our technology. One is through this partnership model where we help you. But on the other hand we have an ability to do it as software as a service. So the goal here is you will be able to come in and there’ll be tools where you can run screens, um, which make easy use of docking and molecular dynamics. On top you would select what you want to do and it would be, um, in a more automated manner. Um, so you don’t need to have a hand-holding from us. And furthermore, we are looking at opening up some of aspects of it, like maybe some aspects of docking and other things to a more general public through tools such as AI assistant or where you can actually you just do a query and it’ll do it for you. And of course, we need to think about which part of it, what is our business proposition there? But there are certain aspects of it which we would love to be available to the general public. So I think that’s where scaling comes in, because you want people in smaller biotechs and universities to have access to part of this technology. Maybe they don’t need to run such a larger screen, but they want to have some of the highest quality tools and make it very easy. So I think of scale as trying to provide enough of a technology to a broader public and number of users learning from them. And then, of course, you need to have the technical scale of, yes, it actually needs to run lots of compute instances in a cloud to do the work. So but that’s behind the scenes on our side, and this is what we spend time on. We can do similar stuff with bioinformatics and workflows and whatnot.
Harry Glorikian: But so that that being the venture guy, right. It begs me to ask the question of, you know, you’ve got this superpower that. Maybe get you to an amazing therapeutic compound. Why wouldn’t you use the technology to take one of those forwards yourself and become a drug company versus make the tools available to someone else? And I’m just wondering, I’m sure you guys have wrestled with this, so I’m just curious how you guys got to where you got to? Or have you, have you gotten to where you’ve gotten to, or is that still being discussed?
Michael Antonov: Well, we are in the early stages and at the same time, this is a very good question. Currently we do some lab work and we do have plans to take certain compounds a certain distance, but we are thinking more of partnerships and a platform at the time being. And quite frankly, this is the dilemma which every biotech platform company has to face. We have really smart scientists who really know what, are excellent at certain domains. And then we’re going to build a platform. And now we’re going to have therapeutics. Now the challenge with therapeutics is it does take a fairly long amount of time. There’s ID, there’s all the wet lab work, and ultimately there’s a clinical stage. So you’re going to need to raise tens upon tens to hundreds of millions of dollars and $1 billion in the later stages, potentially, in this clinical trial. And then you have investors who are coming in. Now, first of all, it’s a probability game because like, I know it’s like 1 in 10 of drugs gets approved. Hopefully with our AI and and the tech it will be I think it’ll be better. But it’s still a chance right. You need to have multiple programs. But at the same time, if an investor is putting all this capital and 90% of the capital is going into a few assets which will make or break the company, their investors, and the board’s focus is not going to be on the technology as much. It’s kind of done once you have your your molecule or whatever the asset is, that is your gold, right? And that creates fragmentation in the industry in the sense that there’s lots of great platform companies which are started and they try to head towards, um, therapeutics, but yet they are siloed from each other. And the technology is not shared as much, which to me as a person who looked at the complexity of these systems and life sciences landscape in general, it feels that it limits collaboration. And we’re not moving as fast. So maybe a kind of a lucky situation that I am in is I can contribute part of my capital to supporting this, which gives a little bit more flexibility and maybe in some, some other startups, to think in a more platform and more open manner. Now, so that means we can take risks to see if certain aspects of tools will get adopted and used broader by the public and experiment with both SaaS and partnership and some of this open business models. And in the end of the day, maybe two or three years from now, you know, it will decide, we want to take some of these assets further down. But right now, a big part of it is trying to figure out how do we have the best possible tech and later on, open it up some parts of it and make it more universal and collaborate more with others? That’s the vision, because I believe that platform company needs to exist, um, in this space.
Harry Glorikian: Yeah, well, it started my career in platforms, so I, I totally agree. Um, but you guys are in a rapidly evolving space. I mean, it’s. There’s a hundred new papers in AI like a day, it seems like it’s just some crazy number. Right? So in the evolution, as it’s moving forward, how do you plan to stay at the forefront of understanding what’s happening, developing it and incorporating it into your platform so that it’s always, the revision is staying up with where everything is going.
Garegin Papaoian: At least on the simulation side, definitely that’s a challenge. And it’s a very rapidly evolving field and it’s easy to fall behind. So one of the things we do is we try to stay current as much as we can with papers. Maybe not. Uh, I don’t know if with all AI papers, but in particular with AI papers that are relevant to drug discovery and simulations and biological modeling. So we part of our time is to actually have discussions of these papers and, and, um, we don’t need to react to maybe most of them, but sometimes some papers come out that does question or does make you question sort of your way of thinking about problems or you realize that, oh, like, here’s an idea that we could maybe use in our algorithms or in our way of approaching things. So we so far I feel that we’ve been able to keep up with that, um, that I in. In drug discovery and biology is a smaller subset of AI. So. So there is more chance that, uh, we, we should be able to do that. But that’s an excellent question. It’s it’s sort of it does feel sometimes your head spinning that, um, that, that it’s almost like every two weeks it’s a new world or every month. So definitely a significant, uh, focus should be on that, not to fall behind.
Harry Glorikian: Yeah. I find myself, like, every once, every week or so, I’m like, no, you can’t do that. Like, wow. You know, and I have to, like, recalibrate myself depending on what I’m looking at. But, um, okay, so looking a little bit into the future. What’s the next big milestone for the company? How do you guys. How are you measuring success? Let’s say or or. Either one of those. What’s the next big milestone? Or how are how are you measuring? Things are moving forward in the right way.
Michael Antonov: I think on a business side, there’s kind of two areas. One is partnerships and customer relationships on drug discovery. Essentially the number and scope. And on the platform side it is users either awareness. If it’s going to be an open tool, which is something will be experimenting with shortly, and direct customer use on the on the platform in terms of SaaS users. So right now, I would say we are still in an early MVP side for some of the life science platform tools. So that’s those are going to be the key metrics.
Harry Glorikian: So if somebody is either a partner or. You know, potential future customers. How do they keep up to date on what’s going on and how do they if they how do they engage you? Is that just going to the website and clicking a button or. Um, just curious.
Michael Antonov: Both. I think clearly having a direct email can, you know, we can email me and can email Garegin, but also, uh, we have, you know, a couple of business development people, uh, on our team is one way. Um, they can read about our technology on our website. Um, and we actually just recently, uh, launched a newsletter which is going to be a monthly. And it has a bit about us, but also a very interesting industry, um, news around our domain. And, uh, so the way to do it is literally go to our website and sign up for the trial for some of the tools and or email us directly about partnerships or any questions they have.
Harry Glorikian: Guys have I have I left anything out? Is there any. Um. I tried to cram everything I could through the through the presentation you sent me, uh, into my brain, but I might have missed something. Is there anything that you guys want to touch or highlight that people should be aware of?
Garegin Papaoian: A shared vision that Michael and I have is to do whole-cell simulations and bring all this technology that, uh, I don’t know at what time scale. Three, five years, ten years, I don’t know. And, uh, but to be able to take a human cell at T cell, B cell, some other cells that are important in disease and do basically a physicochemical, three- rdimensional simulation of that cell. And then you could maybe add drugs and see how the cell responds to it. All these things that you sort of that today you, uh, observe under a microscope. How could we do it in a computer? The answer is that today we cannot for human cells. And because of that, I claim that we still don’t understand biology in a fundamental way. Um, but but when that would, when, when that will be possible, that will be completely revolutionary. And that’s something that’s one of the things that I think drives both of us. And also, uh, the parts of a company that help to basically bring that vision closer.
Michael Antonov: I come from the graphics space, clearly, having done Oculus and the gaming before, so the idea of being able to fly into the cell and keep zooming in up until the point where you see, like the mitochondria and all these spinning things and maybe pause it and speed it up and have it actually be physically accurate in real time or faster or slower seems incredible. So then you can take any measurements on top of it and it would be phenotypically accurate. That would be an incredible vision. So I think as I got into this, um, you know, when you come from more gaming as a software background, it seems into like intuitively like it should be simpler. Why can’t we just simulate? And of course, then you realize there’s quantum physics, there’s all there’s just this full scope of detail as such that we haven’t yet approached it, um, being able to have that compute capacity. But I think that’s what’s so exciting, both about the AI and the data processing capability and instrumentation we’re starting to get now is like, we can get further and further down this path within a decade. Um, and that’s you can simulate a lot of things. You can collect enough data, represent it, and that’s. I am excited to basically be able to see the cell and play with it and really empower scientists with it, um, but also really provide tools which enable others to build those kind of things, whether it’s bioinformaticians, whether it’s around genomic analysis, whether it’s about any data processing to bring it together to make those models. So I think that’s super exciting.
Harry Glorikian: Yeah, I keep hoping that it doesn’t take so long, I know it, I know that it does every time. I think it’s not going to take that long, it takes at least twice as long. But, uh, and every time we think we know something, I realize we we’re just scratching the tip of the iceberg about what else? We don’t know. So, um, it was great having you guys on the show. I mean, uh, I, you know, I obviously I’m going to keep up to date on the progress of the company. It’s already done phenomenally well since inception. And I am going to make the grand assumption that it’s going to continue along that trajectory and continue to do well, uh, and change how your partners are achieving their goals of identifying new therapeutic targets and moving those forward through the through the value chain. So good to have you both on the show.
Michael Antonov: Yeah, it’s a pleasure. Thank you.
Garegin Papaoian: It was really enjoyed the show and the conversation.
Harry Glorikian: Thank you.
Harry Glorikian: That’s it for this week’s episode.
You can find a full transcript of this episode as well as the full archive of episodes of The Harry Glorikian Show and MoneyBall Medicine at our website. Just go to glorikian.com and click on the tab Podcasts.
I’d like to thank our listeners for boosting The Harry Glorikian Show into the top two and a half percent of podcasts globally.
To make sure you’ll never miss an episode, just find the show in your favorite podcast player and hit follow or subscribe.
Don’t forget to leave us a rating and review.
And we always love to hear from listeners on X, where you can find me at hglorikian.
Thanks for listening, stay healthy, and be sure to tune in two weeks from now for our next interview.