Podcast with Shahar Keinan, CEO of Polaris Quantum Biotech
My guest today is Shahar Keinan, co-founder and CEO of Polaris Quantum Biotech, a drug discovery firm that leverages quantum computing to find compounds to treat and cure diseases. We talk about the quantum drug discovery process, their desire to use gate-based quantum computers, whether they can find better COVID therapy, and much more.
THE FULL TRANSCRIPT IS BELOW
Yuval: Hello, Shahar. And thanks for joining me today.
Shahar: Hi, it’s a pleasure.
Yuval: So who are you and what do you do?
Shahar: So my name is Shahar Keinan. I’m originally a computational chemist, and now I’m a co-founder and CEO of Polaris Quantum Biotech.
Yuval: And what does Polaris Quantum Biotech do?
Shahar: Polaris Quantum Biotech uses quantum computing as well as cloud computing, AI, and machine learning, to design novel molecular drugs. Our mission is to cure and treat all disease for all people. And we can do that, because we are using quantum computing. We can do things at a faster and at better scale than other groups that do the same thing.
Yuval: So how would that work? So let’s assume I had a particular disease or a particular virus that I’m trying to target. What do you need to know? And what does the process look like for Polaris?
Shahar: Sure. So you can come to us with a specific disease, or with a specific protein target that is relevant for that disease. And we would also like from you to tell us what are the properties of the future drug that you are interested in? So starting from I’m interested in that protein and I will want to bind to that specific pocket in that protein, but also I am not interested in binding to other similar proteins. So you’re interested in selectivity, but also what other properties you’re interested in, is this drug going to be an brain drug, or something that go through the skin? Is this drug that has to pass the blood band barrier, or never pass the blood brain barrier? So what is the profile of the future drug? And we take the profile of the future drug and the binding pocket of that specific protein.
And together, we build a virtual, bespoke, chemical space that is relevant for that protein, that specific binding pocket. And we search that virtual large chemical space based on the set of properties that you told us that you’re interested in. Okay. And we do it on a quantum computer. The chemical space that we are building is on the order of billions of molecules. What Polaris unique contribution here is, we know how to build that virtual chemical space that is relevant for the protein and translate it to a QUBO formulation. So then we can search it for the properties that you specifically said before that you’re interested. We’re doing a multi object optimization on the computer side of things to find the right molecules that is relevant for that properties and that pocket.
Yuval: But let’s compare it with a classical process. So I’m a physician, I’ve identified a protein target for a virus. I can go, I think, to a lab and say, “Hey, you’ve got a library of 5,000 or 10,000 compounds. Let’s run all of them against my virus, this is how we measure. We see where they bind. We see which have impact.” So isn’t that a simpler process than what you’re proposing?
Shahar: You can even do it better. You can go on the computer and calculate the properties of these 5,000 molecules and test them against your protein, right? And this is something that the pharma industry has been doing for a couple of years already. But the problem is that out of those 5,000 molecules you will probably find a couple of hundreds of molecules that may be relevant for you. And then you will want to improve some of their properties, and then you will improve some of their properties, but then you maybe miss some other properties.
So the whole industry is going in those kind of circles of design, measurements, improvement, and those kind of circles take time and are expensive. What we are saying is instead of doing that couple of hundreds to couple of thousands molecules every time, do that on billions of molecules, find your molecules the first time around, instead of again, going in circles.
Yuval: How long does it take?
Shahar: So the run itself on the quantum computer is extremely fast, okay, we’re talking less than a minute. The preparation takes time. So the project is usually about three months. I cheated a bit, because what happened from the quantum computing is the number of the molecules that we get is quite large. It’s a couple of hundreds of thousands, could be. And then we will use regular computational chemistry tools to go from those couple of hundreds of thousands to less than a hundreds of molecules that you will then synthesize and measure.
So we’re doing this in two steps. The first step is on a quantum computer going from billions to a hundred thousand and then using computational chemistry tools from a hundred thousand to a hundreds of molecules. The hundred molecules come from a very broad chemical space and have all the properties that you’re interested in from the beginning.
Yuval: So when I listen to you and you describe it in present tense, is this something that is working today? Do you have customers that use it?
Shahar: Yes, it is working. We have customers, we published a press release two weeks ago, a collaboration that we’ve done with Phoremost and stay tuned, because later this week, we are going to have another press release with another biotech company that we have signed a contract from, and we also collaborate. We can do collaboration that is more of a fee for service, and we can also do collaboration. So we collaborate with a company called Auransa, that their expertise is on how to find targets. And then we come for those targets, find the molecules, and that’s a good collaboration there. So yeah, we have customers. We have paying customers that work on projects with us.
Yuval: It takes a minute or under a minute on a quantum computer first. This is an annealer, right?
Shahar: Yes, we are working with quantum annealer. We work in the past with Fujitsu digital annealer, and now we’re working with D-Wave on their Advantage system.
Yuval: And is the annealer large enough for you, or you would benefit if it was 10 times larger?
Shahar: So the annealer is large enough for us for the current size of the problems that we’re interested in, which are about a billion molecules. We would be extremely happy if it would be bigger, because then we can create larger chemical spaces. There are ways to work with larger chemical spaces, even now. Divide and conquer mainly, but right now we are working with the advantage system and very happy with it.
Yuval: Ignoring noise for a second. If this was to run on a gate based computer, do you have an estimate of how many gates you would need?
Shahar: I think we will need at least 800 to a thousand, and then we will work with more of a divide and conquer. So we will be able to do a billion, a library of billion molecules, but again, using a divide and conquer. So maybe four quadrants and two quadrants, every time. Right now for us to work on a gate computer, we would be very happy with it. This is something that we would be interested in, because on a gate computer, we will be able to do both sides. So both going from a very large library to a smaller library of a 100,000, and then running those a 100,000 using computational chemistry tool to find the best molecule as well. So solving the Schrodinger equation. So this is where we are. We want to go. This is where our roadmap is, and we are very impatient in waiting for gate computers to reach there.
Yuval: I go back to the one-minute execution on an annealer. How long would it take in your estimate on a simulator, if it was large enough on a multi-GPU, multi-CPU simulator or in a classical computer?
Shahar: So it really will depend on how many nodes you are running. Okay. But I can tell you that our estimation of how much money it’s going to cost right, is about $40,000. And then you can figure out how many minutes this is on a Google Cloud. Okay. So that was our comparison. It is a thousand time more expensive than running it on a quantum computer.
Yuval: So would it be correct that you are building a company that is using quantum, because that’s the best option? Not because you came into this market wishing to solve some big problem in quantum computing, but you’re solving a big problem using quantum computing.
Shahar: Yes. Yes. We are users of quantum computing. We develop, of course, our implementation and using known algorithms. We are excited with these fields that lets us do things that two years ago, we couldn’t do, we are extremely beneficial. This is extremely beneficial for us, because we are just in the right time from both sides, both from the computer side of things, but also from the level of information, that is available on the bio side of things. So genomics, genetics, proteomics, so all of these together make it possible for us to do what we do and to automate quite a lot of what we do and be able to run on a quantum computer.
Yuval: I’m curious about, if you can tell me about the composition of the company, do you have primarily chemists or physicians, do you have primarily quantum computer scientists? How is the company constructed?
Shahar: So we have about half and half people that are on the computational chemistry side and on the engineering side. We are a learning organization. And just now we are starting to figure out that we do need to have a person that has learned how to work on quantum computing. So far, it’s both the engineering side and the computational chemist. Who’ve been working to build our technology and implement it. So we are nine people. We are always looking for good people and interested in talking with people who are interested in joining us.
Yuval: Is your work related or adjacent to the protein folding work by Google?
Shahar: Isn’t it amazing, the work they’ve done? It is so amazing. It is adjacent. So a lot of the targets that we are interested in are targets that we’re so far very difficult to measure their three dimensional structure. And Google is giving us a good starting place to do our simulation. On top of that, it’s opened a whole field of targets that were not available before.
Yuval: You mentioned earlier in our discussion that your goal is to, is very broad, is to cure all diseases. Where do you start? Is there a class of diseases or a class of viruses that you’re particularly focused on right now?
Shahar: So we are building a diverse portfolio of assets, of molecules that will become good drugs. And part of what we want to do right now is to be diverse in that. So this year we’re going to do 20 targets, next year, we’re going to do 50. And after that, we’re going to do a hundred targets per year, develop early stage drug development, working on early stage drug development. What we do now is we either choose diseases that we care about, that we think that there is a market for those that we identify something missing in the market. An interesting subject for us are cases where the market has some solutions, but they don’t work very well. For example, instead of having a small molecule drug, you have a biological drug, and that means an injection instead of a pill. This is a place where there is a lot of work for us, because finding a small molecule is difficult.
You need a small molecule that has very specific properties, and it’s difficult to find those molecules. And there starting from billions of molecules, makes a lot of sense. So we’re looking for things that are missing in the market, diseases that we think are very relevant, but not enough work has been done on those one. Very large segment of the market, for example, is women health that has not been addressed by big pharma so far, although there’s a market for it and enough biology and understanding, this is something that we’re very passionate about. And we of course get targets from our customers, who are it in bringing to us their specific targets and their understanding of biology.
Yuval: I would be remiss if I don’t ask about COVID, right? I mean, mRNA is interesting, but I think there are disadvantages to mRNA vaccines. Is that something that you could help the world with?
Shahar: So we started working on COVID. We had a collaboration with Fujitsu about that specifically. I think these are something that we are really wanting to see where the market is going. There’s a lot of computational work being done there, and it’s not always with the best tools. So we were really worried about just going in there and saying, “Yeah, we’re going to solve those problems.” So we are a bit more careful about what we’re saying about COVID.
Yuval: As we get close to the end of our conversation, once you identify a set of potential molecules or compounds that could work, sounds like there’s still a lot of bench work to do. I mean, is it toxic? What’s the dose? Is it too much? Is it and so on. So what you’re saying is basically identifying promising targets and then there’s a whole sequence of events that needs to happen. Non-computationally bench wise to solve it. Is that correct?
Shahar: Yes. So we are shortening that bench work by starting from better molecules, that the ones that we design, but definitely you need to synthesize those molecules, measure their activity for protein, for cell, maybe in animal models before you can define this as an asset and license it.
Yuval: Well, fingers crossed. Shahar how can people get in touch with you to learn more about your work?
Yuval: Very good. Well, thanks so much for joining me today.
Shahar: Thank you very much. It was a pleasure.