Hype vs. Reality: Why AI Isn't Ready to Make Medicines Yet

Michael Marks: Hi everyone. This is the Tech Surge Deep Tech podcast presented by Celesta Capital.

Each episode we spotlight issues and voices at the intersection of emerging technologies, company building and venture investment. I'm Michael Marks founding managing partner at celesta. If you enjoy Tech Surge, now is the perfect time to hit the like and subscribe button. And while you're at it, leave us a review on your favorite podcast platform.

If you're just discovering us, visit TechSurgePodcast. com to sign up for our newsletter and check out the archive of our past episodes. Today we're talking about a hot topic. The role of artificial intelligence in the creation of medicine. Many are framing AI as the magic. Bullet of Drug Discovery. We've all heard the hype about AI revolutionizing medicine, but what are the realities?

Well, it certainly seems AI has the potential to play a big role in transforming how we develop new medicines, making them better, more quickly, less expensively. Our guest today believes we are way over hyping what AI can currently do to create the therapies we need for the future. Joining us to talk today is Mike Nohealy.

Mike has been a biotech leader for nearly 20 years, serving at the biggest global pharma companies like Amgen and Novartis, as well as leading innovative startups. For example, he was the chief science officer of Generate. He's now the CEO of Prolis Biologics, which is doing very cutting edge work in this space.

So, Mike, thanks for joining us.

Mike Nohaile: Thank you, Michael.

Michael Marks: Okay, before we jump into this debate, let's take a step back and ground ourselves and why this debate really matters. So, Mike, you believe today that we have almost no severe medical conditions for which there are adequate treatments. I think our listeners will be surprised by that.

Obviously, there are treatments out there right now for things like heart disease, diabetes, weight loss, cancer, etc. So, can you explain what you mean by that?

Mike Nohaile: Yeah, so the number one killer of people in the United States and globally is cardiovascular disease. The number two is cancer, um, and I think cancer, everyone can agree that we don't have adequate treatments.

But a lot of people would say, well, why not cardiovascular disease? The two biggest risks are hypertension and dyslipidemia. Don't we have drugs for that? Statins, et cetera. But let's take the reality of hypertension. So we have drugs. If you are well managed at a very good center and you're motivated, we can get almost everybody to goal.

We can reduce that hypertension and, and, Not have them sick. The reality is less than 50 percent of people in the United States are treated to Gall. That's why it's the number one killer. Because people aren't getting there. And you could say, well, is that the drug's fault? But it kind of is. These are pills you have to take every day.

You have to be very compliant with them. You have to do a lot of stuff. And many patients aren't willing to do it. What if you had a medicine you could take once a year? You go get your checkup.

Michael Marks: Yeah.

Mike Nohaile: They give you medicine. And you're done for the year. Okay, I don't know how to do that. Not even working on it right now.

But, but my point is that then we might not be at 50 percent to go. We might be 98 percent to go. And you're losing almost 700, 000 people a year to cardiovascular disease, heart attacks, strokes, et cetera. So I don't think that's adequate. And if you go through the list, lung disease, liver disease, metabolic diseases, none of them, you know, I'll give you one example, hepatitis C.

There you take a 12 week course of pills and we get like 98. 99 percent of people cure it. That's getting pretty close. That's like it. I don't, so we really need better medicines and we need them urgently because we all have ourselves, friends, family that are

Michael Marks: very ill. Okay, so we need better medications.

We'll buy that. How do we do that? Can you talk us through the various waves of pharmaceutical medicine, the idea of wave five therapies and so on?

Mike Nohaile: Yeah, I think if you look at the history of pharmaceuticals, What you're seeing now is a move to more complicated medicines. And the way I analogize this is, we're like cavemen walking around a broken space shuttle, doctors and scientists, and we have a wrench.

And we get to, for, to cure it, to fix it, we get to bang on it once or turn one screw or do something, one thing. And amazingly, occasionally that works and we actually make the patient better. But it's not shocking that maybe it takes two things or three things. And we try combinations of medicines, but I think what we're really moving to are more complicated medicines that can do more than one thing to the patient at once, maybe even read the situation and improve it.

So, and if you look at the history of medication, that's where we're going, right? So a hundred plus years ago, people started to synthesize chemicals. Aspirin was the first thing that was synthesized. That was wave one. Aspirin, Cyanide, Acid. Um, in sort of around 1890. And then eventually, that was the only kind of drug we had, these small molecule pills, aspirin, statins, for about a hundred years.

And then about a hundred years later in the 1990s, well 1980s, people started to develop, um, Uh, different kinds of protein drugs, insulin, things like that, right? Taking advantage of the biotech revolution. Then, um, eventually, uh, people also learned at this point, and it was a little early in the 70s, which I call Wave 2, they learned to do targets.

So initially they were just taking things and giving it to people and see if it made them better. Then eventually our science got better and we started to understand the biology and target. So Wave 2, Wave 3 was these protein drugs. Wave four is what I call the antibody drugs. So about 20 years ago, we learned how to make monoclonal antibodies, and those led to drugs like Keytruda and Humira.

Some of the biggest drugs in the world, like, you know, seven out of the top 10 drugs are antibody drugs today. Um, but what's been happening over the last few years is a move to many more complicated medicines. You're seeing RNA medicines, DNA medicines, but also you're seeing multi specific. So instead of going and doing one thing, they're binding to two things, maybe even three things, and trying them out.

And if you look at people's pipelines, large pharma, you're seeing more and more of these drugs, but they're challenging if you have to make a drug. So, a normal monoclonal finds one target and does one thing, that's hard enough, works like one out of 10 times. If suddenly now you need a monoclonal to do two things or three things, well, each of those has to work and if any one of them falls apart, it falls apart.

And so, your risk goes up, your cost goes up, it gets harder. So, I think We, and I call that wave five, by the way. So wave five medicine are these more complicated medicines. So I think you're seeing again, cell therapies, RNA therapies, more complicated protein therapies, antibodies are kind of protein therapy.

And that's where medicine wants to go because clinically. Maybe we've done the low hanging fruit of the single, you know, turn a single switch, and there's still some, there's still good targets, single targets coming, but you're seeing more and more people wanting to do more than one thing with a medicine.

Those are more complicated, and therefore you need to be faster, better at making those component pieces and putting them together and getting an actual medicine.

Michael Marks: Okay, well, I have a follow up question for that. Before I do, you told, this is just an aside, which kind of cracked me up, and I'm sure everybody will enjoy this.

You told me that you didn't think that aspirin would get approved if it were applied to the FDA today. No way. It's a new medicine. It's a very

Mike Nohaile: dirty medicine. It causes All kinds of stomach bleeds, um, you know, it has bleeding risks. And in fact, it's used as a medication for, you know, you know, preventing clots and things.

But it would have a real tough time, I think, getting through the FDA. It would take a lot of trials to get that thing through.

Michael Marks: Well, I think that's, that's a perfectly good description about how we need. better medicines. Yes. Okay. So drug development is becoming more challenging. You just went through to the wave five and all that, which means more risky and expensive.

And we need a better way of discovering these proteins that could lead to breakthroughs and new drugs. So what are the primary ways we discover these protein binders today and what are their limitations?

Mike Nohaile: So remember, these are protein drugs. So what's a protein, right? I mean, many of us learned it in high school, but just as a quick reminder, right?

So a protein is a linear sequence of amino acids, right? The essential amino acids, and there's 20 of them that we use. There's an infinite number of them, but there's 20 that get used, uh, in our bodies and in other, um, things like us. And they, they form this linear string and they're really cool because, um, you know how you get a mattress and it's all packed in a box and then you open it and it springs and you have a mattress.

Well, imagine instead of that, you had a linear thing and it was in a box and when you took it out and put water in it, it sprung. Into a car and the car worked like it all came together. So these proteins are amazing molecular machines that all come together They do everything in your body everything but many things they Translate your DNA.

They work your muscles. They make your brain work They make they make all the chemicals that are working in your body So they're the workhorse of the body and there's many different kinds One of the things they do though is they bind to things right they find a target and very specifically bind to it So, you know imagine Um, looking at a human body and saying, I want to pick out one very particular part of it.

Well, that's a, that's like a, there's a billion locks. I have to find a key that gets in just the lock that I want. So it's a, it's a, it's a real challenge. And what proteins are good at is they can do that, and there's several ways to make them. One is you evolve them for a couple hundred million years, um, and you get new binders.

And you have a lot of those in the human body. Yeah, it's in the human body. And a lot of your proteins have been evolved through vertebrate evolution and prevertebrate evolution, et cetera, all the way back to like bacteria. Um, it takes a while, right? There are technologies that try to reproduce that.

They don't. Well, we do use technologies like that to improve things, but we don't use them to discover new things, generally. Um, there have been a handful of things that have been discovered that way, but generally it doesn't work super well. So, what scientists discovered almost 40, 30 to 40 years ago, but really perfected about 20, was you take a mouse, You put in a human immune system into it, and we talked about how to do that.

And then you challenge it, and it makes antibodies. And those antibodies are partially human. And what antibodies do is they bind to stuff, right? So you, the vertebrate immune system is the most amazing thing. So you have a non adaptive arm, which most non vertebrates have, which just recognizes bacteria generally.

Yeah. Or viruses. But then we, vertebrates, have an adaptive arm. And that means when you get attacked by something, it will recognize it and make a whole new bind or a whole new system to it. And there's two parts, but the part we're talking about is the B cell part or the antibody part. And so you make these proteins and they bind very specifically only to the thing that's attacking you.

And in fact, that, you know, when you get a cold, you're waiting for the antibodies to come. When we were getting COVID, we were waiting for our antibodies to get made. There's also a cellular response, but, but the antibody part's very important. So you make these very specific binders. And so, well, if you make a specific binder, well, if I can harness that, then I can make a specific binder to the target that I want to modify in disease.

So, for instance, an example is in cancer, um, one of the things cancers often have to do is to fool your own immune system because your own immune system will attack the cancer and suppress it. So, one of the biggest drug in the world today is a drug called Keytruda. It'll be surpassed soon by the weight loss drugs, but today it's the biggest drug in the world.

And what it, what, what it attacks is something called Programmed Death 1, is this factor that gets released that says the immune system by a cancer cell. Don't pay any attention to me. I'm not here. I'm not causing a problem. So then what you do is make an antibody that binds to the PD 1, suppresses it, so then at least in some cancers, the immune system can suddenly see the tumor and attack it.

Right? So that's the basic idea. But how do you get those antibodies? Well, again, they took a mouse. But in a human immune system, and then you challenge it with PD 1 and it makes antibodies to it, then you take those antibodies and you do a lot of work to find, you know, a single sequence, one antibody that turns into a drug that you can make over and over and give to people.

The challenges with that approach are, you know, you have to have this human mouse, you're keeping all these animals, so it's the expense, it takes a lot of time, um, and then the antibodies have some challenges, right? They're sort of a Franken Makeup of partially mouse partially human and I think that causes some biophysical problems and really getting something biophysically good is important for drug That means it injects well and does all the things, you know, you don't want to sit there for 20 minutes injecting yourself Right patients aren't going to do that So anyway, that's how we do it today.

It works reasonably well, but there's real risk to it. Another problem is that even people who say they have fully human antibodies, there are still mouse parts to them. And those parts can cause your own immune system to attack the drug. So you take the drug and your immune system looks at that drug and says that doesn't look like me and it makes an antibody drug.

And that kills drugs. If it gets high enough, there's whole drugs. I lost many drugs when I was doing this at Amgen because the immune system of the patients, maybe it was 20 percent of the patients, but that's enough to kill the drug. And it's sad because you know, it works for a lot of people, but it doesn't work for everybody.

And you can't predict necessarily who's gonna. Cause this reaction, so the drug dies. So there's lots of, it is a good technology. I mean, it's a real triumph of modern science, but there are real limitations.

Michael Marks: Well, and it's also the problem that a drug might work in some populations and not work in other populations.

For sure. And it's very hard, what I hear you saying, what we've talked about in the past, is it's very hard to solve for those problems. with the way drugs are created today.

Mike Nohaile: Yeah, I mean you can spend, you know, um, there's a drug Bocatuzumab, a Pfizer drug, it was a heart drug, and on their final experiment they put 30, 000 patients through it.

It was probably a billion dollar experiment. And they had immunogenicity. And at the end, and they didn't find out. They didn't find out in phase one, they didn't see it in phase two. It wasn't until people were on it for several years, but that killed the drug. So you spent that billion dollars, you spent ten years, and then the drug goes down.

That is Super painful.

Michael Marks: All right. Well, this is a perfect segue into the artificial intelligence discussion because we read all this stuff. We hear all this stuff about how AI is going to come along and solve all these problems and we're going to have, uh, uh, you know, individualized medicine, all that. And I know that you don't believe that's actually the current situation.

So you know, billions of dollars have been poured into companies making a lot of claims and, and, and rightfully people are excited about this stuff. You can understand that. So I want to dig in a little bit and what, what. What we can and can't do with AI, the 2024 Nobel Prize in Chemistry went to the creators of AlphaFold at Google DeepMind, which is a model that uses machine learning to accurately predict a protein's 3D structure.

The award was also given to researcher David Baker, who I know you know, for quote, Computational protein design, unquote. On the surface, of course, this sounds like AI designed or discovered proteins, whole cloth, that can create new medicines. But that isn't what happened, correct? What actually happened?

Mike Nohaile: Yeah, so, I mean, let me be clear. There is a real breakthrough from AI. It is not fake. It does, it does do something important. That's good to know. Um, for proteins. We can comment on anything else, though. It does seem to work for language, too, and images. Um, but, I think what's been oversold is this idea that you press a button and out the other side comes a drug.

So, what I think it works for well, and remember, I was chief scientific officer of a company that pioneered using this. In fact, my technical area is the design of proteins computation. I used to do it by physics, and now I do it by AI. And it does Take existing proteins and make them better. You can vary that sequence, right?

I remember I said it was a long sequence. It can be hundreds and thousands of amino acids. And it will make changes and improve it. It might make 10 changes, 5 changes, 20 changes, and you can get improvements. Um, and that has been done. And when people say there are proteins in the clinic from AI, that's what they mean.

They have been improved. They've been improved once.

Michael Marks: They haven't been created from scratch.

Mike Nohaile: They have not been created from scratch. So you take a reference molecule that already does what you want, partially, and maybe make it bind more tightly, you improve the function. And that's a big deal. Don't get me wrong.

And in fact, you know, it's a big enough deal that at Prellis we're, we're using that as part of what we do. But the real question is how to get that seed in the first place to a new target, right? So, and there's two reasons. Some targets haven't been hit by antibodies at all and some targets have been hit by your competitors and they own the AIP to that seed and you can't go forward and make the medicine you want because It's not your molecule.

So, um, so basically, you know, it is a big advance. I think it is important, but I think it's been oversold because people use the word novel. And if by novel you mean the sequence has never been seen before, that's true. If by novel you mean it's hit a whole new target in a way no one's ever seen before, that, we haven't seen yet.

Uh, and I don't think we're that close to it either. Uh, on the first part, I'm completely rock solid, right? That's just, that's just the facts. You can look at the pipeline, you can look at it. Now I'm moving to slightly more speculative territory because somebody could say, okay, that's great, Mike, but you know, computers are getting faster, data's getting better, et cetera, et cetera.

Algorithms are getting better. So next year we'll have it. I don't think so. Um, and the reason that I think it's further away than people think, and I think decades, is because it's a data problem. So if you think about an antibody, the part of an antibody is many, many hundreds of thousands of amino acids, but the part that actually binds, the part that buries, is 60 amino acids, roughly.

It's called the complementary determining region, CDR. So if you think about how many different antibodies you can make, there are 20 different amino acids, 60 positions. 20 to 60th. That's a big number. That's more atoms than are in like a trillion, trillion galaxies. Individual atoms, right? Right. And so, what the technology does, the way I think about it, is it's like you're standing in a, in a, looking, um, for your keys under a light, right?

That old thing, that you're looking for keys and you look where the light is? Right. And it used to be, before AI, we had a very narrow light tone. You could look just right by using the alignment. Um, hidden Markov chains, alignment kind of stuff. And what the AI has done is it's dramatically expanded the size of that.

search around what you already have, but that's, you know, so let's say initially you could look at a little, like a foot radius. Now you can look a hundred feet radius, a thousand feet radius, way more, a mile radius. The problem is the size of the thing you're looking at is the solar system. And so maybe there's a thousand places and you have a thousand spotlights and you've made them vastly bigger, but there's still a lot of black, right?

And, uh, but I mean like 99. 999999999 percent black in the sequence space. And so as you open that up and get more data and find things in each of those spaces, you can then use the AI to help fill in the circle around it. But many things we've never hit, and the only way we know to hit them is biologically, not computationally.

So I think it's going to be a while until we. have enough data that you can do that push button thing. And by a while again, I think.

Michael Marks: Well, it's interesting. So I'm going to drill down on that a little bit. Based on a 2023 article in Nature, AlphaFold, the Google product, predicted the structure of every protein known to science at the time.

Can that be true? Is that a true statement?

Mike Nohaile: Yeah. So go back to the Nobel Prize. It was given for two things. One half was given for what's called the protein folding problem. That's given a particular sequence, what does it fold to? And that's a real breakthrough and it works pretty well. Now. The statistics are a little bit deceptive because it's best at the parts that you already kind of knew, the solid parts, the parts, the action parts of the molecule, the loops and the other things, it does a little less well, but it's a real advance.

And for certain scientific applications, it's really good. There are other applications where, You still need to go do the scientific works, crystallography, et cetera, et cetera. But, but it's a real advance. The other half, but that's not our problem. Our problem is the other half. I have a target I want to hit.

Give me a sequence that hits it. Um, and that part, again, my assertion is we can find things that work if we have a good seed. But in fact, the first part can be true. That alpha fold folds every protein roughly to its,

Michael Marks: Right.

Mike Nohaile: Without implying that then I can find a sequence that gives me what I want. Um, and so, again, I think even on that side there's been a real advance.

Because the ability to look at more sequences around a seed and find things that work better, that's a big advance. Don't get me wrong. But I think, unfortunately, because it's quite complicated, it hasn't been well understood, especially when you read the New York Times article or something. What it's saying is not well understood.

For practitioners in the field, it's a big deal, but it's not the solution. It is a piece of a solution.

Michael Marks:  So really, these are improved proteins, but not novel ones.

Mike Nohaile: Not novel. Again, so let's say you had a new target that, or a new part of a protein, like, you know, so take that PD 1, right? It's not too hard to find a way to block PD 1.

As I said, PD 1 you use to tell the immune system, attack here. Well, what if you had an autoimmune disease? It turns out there's a particular part of PD 1 you hit that might turn it on. And you might want that. If you had an autoimmune disease, you might want to turn it on and say, don't attack here. Um, but if you don't have an initial seed, the AI is not going to find that.

It just, because it's a whole new part of PD 1, a whole new antibody that binds to a new part. That's not, it's not showing the ability to do that. So someone needs to biologically create that. With the mouse or with another technique, which I think we'll talk about in a minute. And once you have that, then you can use AI to make it better and make new versions, et cetera.

But until you get that initial one. I see no signs, I've talked to many other practitioners, I think if you go and talk to the heads of research at major biopharma companies, which I have, they feel it was oversold to them a couple of years ago, and they are a little down on it these days. Well, that's

Michael Marks: not too uncommon with new technology.

People try it, it doesn't work, they don't want to come back to it, they don't want to hear about it again, so I assume we're in that mode here. Why do you think the industry and the media It drives me crazy in almost every industry and here, you know, I read about, you know, AI is so good, we're going to have individualized medicines and you're debunking that entirely.

So why, why do they say it?

Mike Nohaile: I mean, I think it makes good press and, you know, the scientists are incented to, to hype their results and that's fine. I don't have any problem with that. Um, and I think honestly, the other thing is there's a big, The big difference between getting a nature paper, you know, so in, in biological, biological research that the three big journals are nature, science, and cell, and getting a paper in those journals is very hard and very prestigious, gets you a 10 year, it gets you lots of things.

Um, that's a far away from a drug, right? Even if you have something interesting, there are years of work. It's, it always amuses me because. Sometimes you see, uh, political statements that everything that biopharma does, NIH did. I mean, NIH does very important work. They fund very important work. That's the first couple of years of the work.

And then there's 12 more years of work that happen. Billions of dollars that go in, and it's not easy. And so I also think it's just hard to explain all that. Not many people understand it unless you're in the field. And even some of the scientists that are making the claims, they're just not drug people.

And that's not a criticism. They're brilliant, brilliant people, but they don't understand it. know all the steps, so they see they're holding a sequence, and they're like, this binds, it works pretty well. Well, that's like one tenth of what you need to have a drug.

Michael Marks: Well, my favorite quote from you, which I'm sure the listeners will appreciate, is you say, if AI can create drugs, where are they?

Mike Nohaile: Yeah, that's, so I've given you a theoretical argument about why I think we're far away, but my count, my other counter is, these technologies are now several years old, former versions of them are a decade old. I don't see any drugs. I mean, people say they're novel drugs, but if you dig into it, you will find there was a seed and what was happened was it was varied around that initial thing.

I know of no cases where an antibody has been made de novo. Um, and the couple of things that people say have been de novo are these very small things that might work in a couple of cases.

Michael Marks: Yeah.

Mike Nohaile: But they're not going to be a general solution to the drugs that you need. Um, there's some people that disagree with that.

Uh, you know, uh, and say, no, no, those will lead to it. Okay, prove it. I mean, I'm, I'm, I would be so excited if we had a push button thing, cause I just want to make great drugs and I want to cure everything, but I just don't think we're there.

Michael Marks: Okay. So we understand that AI alone isn't sufficient to discover new medicines.

We just talked about that. But we also know we need a better way of discovering the proteins necessary for more complex medications today. Can you talk some about that, what the missing pieces are, what are the things you're trying to do to solve that problem?

Mike Nohaile: Yeah, I mean, as I said, there are real limitations for the existing technology in terms of how fast it is, in terms of the quality of what you get, in terms of the risk.

So, Imagine you're doing a drug that you want to bind to three things, not just PD 1. You know, one of the biggest results recently was a PD 1 and a VEGF put together. So, it binds to two targets at once. And it's the first thing that's done better than a PD 1 against certain kinds of cancer.

Michael Marks: Okay.

Mike Nohaile: But now you have to get two antibodies.

And let's say you have a 70 percent chance of getting each antibody. And they're independent. There's no, they're not correlated. So suddenly you have a slightly less than 50 percent chance of getting the whole molecule. So that adds a lot of risk. By the way, you have to get both. That's double the cost.

You have to paste them together, which actually accelerates the risk. So it's worse than half. So all those things go into it. And so if you sit in the companies making these decisions, on the one hand, you're pushed to go that way because. That's where the clinical science wants to go. On the other hand, it adds time, cost, risk to the proposition.

So you'd like to lower that. So if you had a technology that was faster, uh, if it's faster, probably gonna be cheaper and higher quality, i. e. lower risk and better when you got there. Antibodies are the best way to do it, but if you put them in mice, they have these limitations. So how do you get around it?

Well, one way is, what if we could make truly 100 percent human antibodies out of humans? Well, that's unethical. You can't actually inject humans and do the things it takes. Because you, if I injected you with PD 1, um, besides probably causing side effects like making the immune system suppressed, uh, I wouldn't get any antibodies because your, your body would say, well, that's just me.

I don't, so you have all these layers of systems that prevent you from making antibodies to yourself. And in fact, when they go wrong, you get an autoimmune disease, right? That's when you get lupus or you get RA. It's when you start attacking yourself. So we said, okay, um, but what if we could trick the human cells into?

Doing it and making those antibodies anyway, because then they would be a hundred percent human. There's nothing not human about them. They would be fully human. So what we've done is develop the technology that allows you to do exactly that. It turns out if you look at your blood, you have all the cells.

your immune cells in your blood. You have B cells, T cells, antigen presenting cells, the whole complement. So we take human blood and we extract those cells and sort them, B cells, T cells. And that's good. But if you just have the cells, that doesn't do it for you. So then we use a two photon laser system to carve out the appropriate space for these.

And we put T's and B's and all the places they belong. And we make an organoid, right? So it's like a little mini organ. And the organ in your body that does a lot of this is called a lymph node, right? You get them, they swell under your neck or in your groin or under your armpits when sometimes when you're sick, if you're really sick, that's because the bacteria and your immune cells are fighting there.

Um, and that's a place where a lot of your antibodies are made. Not exclusively, but one of the places where, that does a really good job of making antibodies. So we're trying to, we have recapitulated that. Um, it's not a true organ, but it's an organoid. And so, and it's fully human. It's one, there's nothing that's not human in there.

Uh, we even for a while we were using some chemical components that came from other places and we're like, no, no, 100 percent human. Um, so fully human, the cells are there, they're plated, and we can make a lot of these lymph node organoids in these special surfaces that we create, and they're three dimensional because that's what the lasers do, they build up a three dimensional surface and make it more like an organ than just cells just sitting in a dish.

And then what you do is you need a set of tricks to add the thing you want to target, it's called an antigen, so let's say you put in PD 1, and then how do you get it to actually recognize that as foreign? It's human, it should just ignore it, right? There's lots of systems that your body has to ignore it.

So we came with a bunch of tricks to get the immune system to recognize it as foreign. Um, and then the, the system starts churning out antibodies because it recognizes it's foreign. And then we use a common technique to pull those antibodies out, figure out what they are. And then we get what's called a library of antibodies.

You know, thousands and millions of different antibodies. Uh, and then at that point, you're in regular drug discovery. right? Like that's something everybody recognizes at that point. Now, one twist we have is we are using the AI because it is valuable, right? So what's traditionally been done up to this point is you use a different kind of biology to improve them from that point because they're never good enough to be, I shouldn't say never, almost never good enough to be drugs right out of the gate.

You need to keep tweaking them and changing that sequence. So we are using. You know, the best modern AI to do that. And that's the best use of AI today. I think that's the best. And it's, it's significant. I mean, really, I am super psyched about the Nobel Prize. I just think, as always, you should be really clear what they deserve a Nobel Prize for.

They did deserve a Nobel Prize. Right. I just think a lot of people walk away thinking it does something it doesn't do. Um, but we use it. It works extremely well. You can very rapidly, cheaply, and most importantly, high in high quality way, make a better set of antibodies. And you want to set because you're losing all the way along.

Because as I said, you know, people get fooled. They think, well, I have something that binds in a dish and it works really well in my little assay in a lab. It's a drug. I don't know. That drug now has to bind in a much more complicated environment. Your manufacturing system has to be able to, we lose drugs because you can't express them all.

You just don't make a lot. So the manufacturing people say, oh, I can't make this or they can't purify it.

Michael Marks: Right.

Mike Nohaile: Or they can't make it soluble, right? Nobody wants to sit there for 20 minutes injecting themselves at home. That's not going to work. Now, for cancer, that's not a big deal because people sit in infusion

Mike Nohaile: for weight loss, for cardiovascular, they're not going to sit there for 20 minutes and inject themselves. be painful, by the way, too, if you don't get this right. So there's all these things that drug has to do, and the AI can be very helpful at rapidly, cheaply, in a high quality way, delivering, you know, moving it towards.

Michael Marks: And are you yourselves at Pellis advancing the AI? Is this a matter of the data you have? Are you doing work on that? Or are you just using.

Mike Nohaile:  That is our ambition right now. Um. Google DeepMind released a lot of the output of the world and the academic community has rapidly improved them. Google. Demis. Uh, it was very helpful.

Michael Marks: Good.

Mike Nohaile: Um, and, uh, but we think we have data that will allow us over time to improve, particularly in areas like the biophysical properties of drugs. Yeah. Um, uh, right now with this, where the most data is improving binding. Um, but, you know, over time you want to improve these other properties as well, because again, you can have a perfect binder, functionally works great, if it doesn't, if it's not soluble enough to fit in an injection, or it doesn't express in a manufacturing system, it's still not a drug, right?

Like, you know, it just, it just isn't because at the end, we have to deliver a product that people can actually use and get benefit from, and that's just a different level.

Michael Marks: This is incredibly exciting. And so can you tell us how far. You've moved this technology forward at Prelis and also maybe some of the targets you're working on?

Mike Nohaile: Sure. So, um, at this point we have, uh, shown that we can generate human antibodies over more than a dozen targets. Um, that making that initial B cells, making the antibodies, purifying them, showing that they bind and, and that's the traditional way to do it. Um, we are now moving them ahead towards the drug like properties.

‍Michael Marks: And if I can, if I can just interject. Please. When you talk about 12 Drugs, can you relate that to how the pharma industry is?

‍Mike Nohaile: Yeah, I mean, you know, what you'll see is a large pharma company, you know, the largest pharma companies in the world try to advance, you know, call it 8 to 12 a year. And you're doing 12.

Michael Marks: Yeah. And you're not exactly a major pharmaceutical company.

Mike Nohaile: We are not. We are doing this with 50 people right now. Um, yeah, it's very efficient. I mean, these organoids, you know, the mice are a little fragile. And, you know, you can only use so many mice and all that kind of stuff. It's very easy for us to make.

Well, they come in plates of a hundred and the other thing that distinguishes that if you're, when you do the mouse, you sort of have one human immune system in your mouse. Now, Regeneron's human immune system is different than Amgen's, but they sort of have one. I mean, I start with whole blood. I can do 20 donors.

I can, and it turns out that humans are really different in their immune system and their responses. So, that drives up the diversity and you want high diversity because you want lots of different solutions because you don't know which one's not going to express, but if you have 10 good solutions, Okay, one doesn't express, it doesn't matter, I've got nine more.‍

Michael Marks: So if I understand this right, if you have a mouse, your Regeneron, you have your own mouse, you, you will get one solution.

‍Mike Nohaile: Well, you get few. And if it works well, you may get many, but you may have to work hard to get the many.

‍Michael Marks: But in, but with the Prolis system, you can get dozens and dozens. Yeah, we get, we get

Mike Nohaile: a lot of solutions right out of the gates because, you know, humans are quite diverse.

Also, the, the, the immune system of a human is more complicated than the immune system of a mouse and you get more solutions out of one. But the real advantage is you can do 10, you can do 20. And all I have to do is buy a blood and, and pull it out. You know, it's just a, the mouse is a much harder problem, right?

Again, it's amazing technology. I just think this is, this is, uh, this is an easier, you know, honestly, ours is easier.

Michael Marks: All right. So that's a good lead into what are you working on?

Mike Nohaile: So, um. We're working on a, a, a few things. So we're working on some of the classic targets like, uh, PD one, not that the world needs another PD one, but it becomes a component of, and that's a, that's a cancer drug I mentioned earlier.

Mm-hmm . But it is a component of many of these bispecifics and multi specifics. You want a PD one arm and then other arm. So we are, we're working on that. We're we, of course, like everybody else are working in the space of, um, obesity and metabolic disease, which is. Yeah, I'm fascinated by this.

Michael Marks: You can drill down on that if you would.

Mike Nohaile: So look, the current obesity drugs work extremely well. We're seeing people getting 15, 20, 25 plus percent weight loss. Um, there's always room for more. There's some people that need even more, but there's a lot of other problems. The side effects of these drugs, the muscle. loss that people see with this.

So we're working, I don't want to get super specific on exactly the targets we're doing, but we think we have angles on more complicated molecules that would address the muscle loss, potentially have lower side effects, and certainly have some better characteristics in terms of, um, administration. So, right now we're in the sort of wave one, the Ozempics, the Wegovys.

And what's coming is a whole wave of wave twos, and there are, you know, a dozen different things people are trying to do. And, you know, I think because we can make a better drug, we think we can participate in that. We have, we have two or three targets, or two, you know, three. I think three, the fourth we may do.

Uh, where we think we have a better mousetrap and we think we can get there pretty fast with a really good drug that, you know, again, has better muscle characteristics, better, better side effect characteristics and better administration characteristics. And, you know, hopefully these drugs will be with us for a long time.

I, I was a little bit initially skeptical of this whole class, uh, but you are seeing some amazing results for people like people that lose, you know, 20, 30, 40 pounds. I mean, it is making a profound impact on their health. It is. So. you know, we need better drugs. We need, we should, this people, I mean, you know, you can always say people should just not eat so much and they should exercise more, but they're just not going to do that.

Right. Right. So we've tried.

Michael Marks: Well, it's fascinating going back to something that where we started this discussion is that there's almost no really good medicines, which is I found shocking until I heard the explanation. I think our, our listeners will hear the same. And so you're always talking about how there's just hundreds or thousands of opportunities to, to make for better medicines.

Yeah. But what are the hurdles left for you? I mean, how far along are you? Is it proven? What, what, what happens next? Yeah. I mean,

Mike Nohaile: I, I feel very comfortable. We can routinely make antibodies to most everything we've tried at this point. I assume we'll find some targets that are challenging. Um, You know, but so far we've tried some very hard targets including, uh, GPCR, which for those that are in the know is a hard class of targets for antibodies, but we have hit a GPCR very quickly.

And when you say you hit it, it means you've created an antibody. We've created an initial set of antibodies to it. That looks promising. That look promising. You know, at this point, we're still a year away from having real drugs that are ready to go to the clinic, but that's what we're racing to do. Um, you know, the hurdles are, You know, it's an execution hurdle.

Making drugs is hard, so we've tried to hire some very experienced people that have done it many, many times. It's always interesting to get people that see the vision, but have the reality of what it takes to, to actually make a drug, but I think we have a number of those. Our CSO, Les Miranda, is amazing.

Our head of biology, Clarence Hale, is amazing. Um, so it's that execution challenge in making stuff. It's, it's getting the scale, and, and, you know, there's things like, Not that many people have cracked the AI thing. We're still, we have it working, but you know, getting the, what we call it wet lab, dry lab.

That's the common thing. The dry lab is the computational and the wet lab is the, literally the wet stuff we put in test tubes. You know, getting all that stuff to work, automating it, making it fast, making it reliable. You know, our, our biggest problem still is operator error. People put a little too much of this in, not enough of that in.

So, you know, as we automate more, I mean, yes, automation is cheaper, faster, but honestly, it's a quality issue. Right, so there's all those sorts of things and then our other big challenge when you have technology like this, this is called a platform technology, it makes the drugs. But the other big piece of it is picking the right targets.

So we are hiring very good biologists who are experts in these target areas to help us choose the right targets and You know, those are probably the two things that keep me up the most at night. Uh, I have a great set of investors, so I don't have to worry too, too much. Thank you, Michael, about investment and things like that.

But, um, but, you know, am I messing up the execution? Are we getting the right people in the right places? And are we picking the right targets? I feel very confident if we get those two things right, we will produce a lot of really good drugs.

Michael Marks: Well, look, it's, it's completely fascinating. I know that I'm speaking for everyone who's listening and myself in this journey as I've been on with you is like what a great problem to be solving.

I mean, so many, you know, people are living longer, we're having all this stuff, but we know that, we know that there's, there are many diseases that we can't really cure. And I know that we're, we're, I'm speaking for everybody. I'll hope you're successful at that. We do. As a, as a final. comment, I'll make a personal comment.

I mean, you know, we at Celeste invest in all kinds of technology companies and, you know, invest in really good entrepreneurs and, you know, it doesn't get talked about that much, but the thing is, the best thing we can possibly do is to fund entrepreneurs like yourselves, that not only will create great companies and have a good financial return, but it'll make the world a better place.

And I, for one, can't think of a company better than this one for solving both. So I really appreciate what you're doing. I

Mike Nohaile: appreciate it. And yeah, that's the way we feel. I mean, obviously we're all competitors, but I get so excited when someone gets a new drug that works. Like I, I have a lot of admiration for Lilian Novo for this new class of obesity drugs.

It is helping a lot of people. You can help a lot of people. You know, I've, I've been privileged in my career to work on some drugs. They're helping a lot of people. I get my small part right past the thousands. Yeah. But, um, but you feel really good. You meet somebody that's actually benefited from the drug and you feel so good about doing it.

So, yeah. I mean, yeah, let's cure everything. I, I'd love it that, you know, you could tell a parent there, here's your new baby. He or she's going to live to 95 in good health. Right. Right. Um, that would be, that would be amazing.

Michael Marks: Well, we are thrilled to be invested in your company and I thank you very much for taking the time today.

This is fascinating. Thank you, Michael. And good luck to you.

Mike Nohaile: Thank you.

Michael Marks: Thanks for tuning in to the Tech Surge podcast from Celesta Capital. If you enjoyed this episode, feel free to share it, subscribe, or leave a review on your favorite podcast platform. We'll be back every two weeks with more insights and discussions of all things deep tech.