Directed Evolution of Antibodies with Doug Chapnick

Welcome to the Axial Podcast. Axial is an early-stage investment firm based in San Francisco. We partner with great founders and inventors investing in early-stage life science companies, often when they are no more than an idea. Axial is fanatical about helping the rare inventor who is compelled to build their own enduring business. Hey Doug, thanks for doing this. Yeah, it'd be great to just chat up about BioLumix, your story, and some of the lessons you've picked up over the years from translating an invention you made into a company. Maybe to start off, just who you are and what BioLumix does.

Yeah. So, Josh, thanks for having me here on your podcast. Yeah. So BioLumix is a very different company than I see other biotech companies, not just at our stage, but other stages. We are doing things a little differently. So we came out of a DARPA project at University of Colorado. So let me rewind all the way to the beginning. So in 2012, I got a PhD from University of Colorado. And at the time, I was very much bleak on how research is done. And it's basically like, no one knows anything. I was echoing, I don't know if you saw recently, Roger Pullmader said, it's a miracle that we've made any drugs. We don't know anything about cells. And that's basically what I felt like in 2012. So not much has changed. And ultimately at that time I said, well, am I going to go do a postdoc where I try to learn how cancer cells work, even though apparently no one can help me learn how cancer cells work to a degree which we can predict things, we can make drugs, things like that. So the relative success of building those models is very, very low. And so this opportunity arose where DARPA— DARPA is basically like the agency that lets you run wild with a bunch of money and go after things that no one else will fund. And so what they had was this project called Rapid Threat Assessment. It was this idea that you could take existing technologies and you could take a bioterrorism threat. It could be an antibody. It could be a small molecule. It could be mustard gas stimulants. And you can figure out how they work without any literature. So no models. You just make measurements and you piece together a model totally based on data, which is basically what the scientific method is supposed to be. But it's not what we do, right? We take literature, we try to build incrementally, and we try to refine models as we go. We don't try to do it all in one shot in 30 days, which is what that project was pitching. And I heard that, and my friend was in charge of the grant at University of Colorado, and I basically said, can I do this? Can I be part of this? So I did a postdoc for 7 years in this project. And ultimately what the project told me was you can do a data-driven walk through hypotheses very quickly if you can do a lot of tests at once. So basically that's the story of drug screening, right? Small molecules are screened in these multi-well plates., and people look for the well where the activity is the highest. They had some way to measure if they're going up or down the landscape. And so then it's just a matter of how many drugs can you put into that assay, and then you say, aha, I found the best one, right? Well, in the protein world, that doesn't exist. And so phage display finds binders, but it doesn't measure activity. And so there really isn't an activity-based giant high-throughput screen. Screen for antibodies. And so BioLumix came out of this DARPA technology where we were programming single cells to have genetically encoded biosensors so we could see which cell had what activity. So the eureka moment in that project, year 6-ish, was, oh my God, if we can make one cell, one test, we can see already which cells have activity. It's just a matter of how do we get that cell to be the only cell that that secretes an antibody to itself, and the next cell does another variant of an antibody. And so that's basically what BioLumix started on, the hypothesis that, oh, in this project, we are very good at building tech, very good at stitching existing tech together. We can solve this problem going forward. So when the project winded down, we basically started with just that idea. Now we know how to measure single cells, what's going on using light and microscopes. Our real first couple of years was how do we make self-drugging cells at scale where each cell has an antibody test packaged up locally in a sea of cells so that you could do one 96-well plate. So 96 tests normally would now become a 10 million test plate. And so, so we started in 2019 going down that road of of doing directed evolution in human cells.

Was it obvious for BioLogix to focus on antibodies initially, or was that—

Yeah, there was actually like somewhat of a plan. I wouldn't say it was, um, we weren't shouting it from the top of mountaintops, the whole plan, but, um, at the beginning we were saying, well, we could make assays the same way. So another protein that you put into cells So the genetically encoded FRET biosensors, they're not antibodies, but they allow you to build a giant toolbox where if you're good at making a big pot of these biosensing cells, then fundamentally you could transition to have an assay that works for directed evolution for any target. And so a good example is we had, we set out to make the world's most quantitative EGFR assay. So that's one cancer drug target.. And just using light, we can tell now today with like 99% confidence whether the EGF receptor is on or off in any given cell. So at the single-cell resolution, we can make a call that's very confident whether a drug is working on that cell or not. And so that was like stage 1. Can you make the measurements to do the evolution? And stage 2 was how do you get the self-programming cells? So when we were at stage 1 and we were doing assays only, we were basically using these assays for other biotech companies that needed assays for targets. I think what most people are feeling is that the assay world is basically Eurofins and a couple other companies, but they're very quantitative on the well scale on some level, or you need lots of cells to get competence, but they're very poor at the single-cell level. And so we had no tools we could purchase off the shelf to do this whole project, this whole company. We had to build every tool piece by piece, starting with the assays that we do directed evolution with. So we did drug screening with people. We made some cash through revenue, and we learned a lot that people, they really need assays, but they don't value them as much as drugs. And so it's a very hard road to build an assay-based company. We actually tried to do it and put it to the test, and it was very clear when we got our antibody self-drugging cells stage done. And so now we could take those assays and one cell would be one test, and we can program one antibody into every cell, and we could do a million tests in a plate or 10 million tests in a plate. Then we were like, oh, time to raise. Because now people care about drugs. They don't care about assays. We're not headed towards glory with the assay route, but we built our toolbox that we needed to do the next stage. And so it's that story of sort of helping other people have assays. It's sort of a temporary business plan. And then you do the pivot. In our case, the pivot was in sight. We just didn't know when we would get it because we basically didn't at that time know when we would solve each problem one by one. We just knew if we have revenue, we could keep going for every year.

And I think that's kind of the, ultimately therapeutic development is a, I see it as an economic black hole in a good way where it's just you're, if you're building anything interesting in biology, making it therapeutic is probably the most valuable impactful thing you do. But those early years seemed of doing services. Do you think they were formative for BioLytics?

Oh yeah.

Do they help maybe create initial culture, or do they help with defining what's important to focus on therapeutically? Like those initial years of doing services, how do you think that helped the company?

Yeah, so, so, so it's not just the initial years of doing services, it's the years concept. So slowness, the time range that we worked in. So we started in 2019, we basically did our first VC backing in 2023, and in that whole time we were inventing, right? And so when I was at University of Colorado, I'm sure I was similar to most people that are in universities. You're like a little surly, you're a little like working on your own thing, kind of taking advice, but also kind of being like, I know what I'm doing. Don't bother me. And so like when I became an entrepreneur, I had to shed those things. That's basically a characteristic of a bad leader. Those are characteristics of people that don't support other people. And I'm not, I can only speak for myself. I don't wanna, so, but it's what I had at the time and I had to shed that. And I had two co-founders, so Jeremy Jacobson and Ted Key. Jeremy came out of the DARPA project, so he is surly too. He came out of university. Ted, not so much, very upbeat person. And so in those times, we were doing all this inventing, right? And we also were always near having no cash. And so there's generally a wall in front of us that just keeps moving forward every time we sell a contract or whatever. But in that hardship, in those hard times where we're lurking in the unknown trying to build this technology. So it's everything from how do you clone 1,000-fold more than most people can clone? How do you make giant libraries with giant sizes, which is one of the big limitations for cloning? We had to solve all these different problems. Not just like biosensors, it's also hardware. We built single-cell pickers. So Jeremy machines metal, programs hearts, circuit boards to basically build robots with the world's tiniest needles. Our tech is sort of this synthetic biology. It's this single-cell picker. So these pickers that have needles the size of human hairs that are guided by computer vision and vacuum cells off of plates. So when we find rare cells that have an antibody we're interested in, the cell picker goes and vacuums it. And that was basically invented fully and patented as an idea, or filed as an idea, all by Jeremy. Right. So he just sat down and said, this is the best way we should do it. And then he started building it. So he's building that. I'm working on this synthetic biology thing with Ted, trying to get these self-drugging cells to work. We're also trying to crack this cloning thing at scale at the same time. And we're really like feeding off of each other's strengths and insecurities, right? So we're pushing each other's insecurities down. So we're all kind of growing as this is happening. And there's this wall in front of us that's sort of driving it. It's either not going to work if we're not efficiently working together and we're going to fail and all these years would be a waste. Or, you know, I get less— me and Jeremy get less surly and we all learn a little bit more about the the first principles that maybe we didn't appreciate that are the source of invention, but I don't think are really like what people work on in academia that much. They mostly try to build models.

I think Appcelera is one of these other examples, but they didn't raise institutional backing until like 5 years after founding. Based in Vancouver, just kind of gave them the space to establish a platform and kind of get out of the academic mindset. And so for BioLogix, how did you, for you, Jeremy, everyone on the team, how did you, what resources did you seek out? What kind of expertise did you get to become a better leader? Also, the one thing, how did you not get misdirected?

That's good.

Yeah. As a VC, I see, I know a lot of scientists, they get horrible advice and they take it. How did you avoid the bad advice?

As well, a lot of thinking is the answer to that last one. So that's a true thing. I think across not just entrepreneurs and that whole landscape, people love to give advice. And if you ask for advice from enough people, the burden will be on you to figure out what's the best advice. Even if you go after an expert, you might catch them on the wrong Tuesday and they might have a totally different opinion if you catch them another time on the same topic. And there's some reliable— I'm not saying it's all unreliable, but the burden is on you to do the panning through the advice. And the best way— so for me, how I dealt with having conflicting advice was put some nuggets to test, to the test, and say, well, if that's true, then we'll do this and we'll see that. And so you sort of, you sort of get in this mindset, well, advice is great, but trial and error is how you weed through it. And the key is to sort of do the least amount to find your way. And so, so, and that's really what the DARPA project actually ingrained into me and Jeremy is like this whole idea of like there's 1,000 possibilities and you've got 3 tickets. And so how do you think through maybe 3 tests that would tell you which of the 1,000 are the best way to go through forward? And so it's basically a game theory way to live. So we sought out advisors and some told us good advice and some told us advice that didn't pan out. But ultimately what we tried to do was, was see where data showed us the path was the most fruitful. That's basically the concept of a drug screen. That's the concept of the scientific method that basically no one follows. But it's this idea that you are a slave to the data. The model doesn't— isn't what guides you. The data guides you. The model is how you talk to your friends about what you're doing. But ultimately in our case, we just basically were in here, we were actually in this lab, this is 1,000 square feet, that's now an office, two of our offices, but it was our original lab. So there was 5 of us in this tiny room just trying 10 ideas every day. And then we would regroup and be like, okay, there is a glimmer that this is the right way, and we have all these advisors and we bounce them off and we'd say, What should we do if we think this is the right way to clone? And on some level, we just basically trial and errored our way to be able to clone millions of genes in one reaction, which I think right now, as far as I know, we're the only people that can do that for giant mammalian vectors. So it's like this really the story of trial and error game theory of what the trials should be and who should you balance what data off of. And it's not the case that you, you just like splatter data in front of somebody and say what to do. That's the mistake that we started with where we were like, we don't know what to do. You have to do take that burden and say, I've curated all of this. I think there are 4 choices. What do you think? You know, so, so it's a very much, you can't ask the advisor guide you 100%, you have to sort of guide them to guide you. And it's this iterative loop of advising that got us forward. So we have a number of advisors today that we're very proud of and that basically are used for very different purposes in the company.

So on this theme of directed evolution, how did you get to therapeutics? What I can say is being an investor for some time, maybe 7 years, almost every biosensor company I've interacted with got caught up in the trendiness of water reclamation or some maybe industrial climate market. How did you not get caught up in maybe the trendiness of some new market and just say therapeutics is the best place for our technology?

Yeah.

Was there some sort of interaction you had with pharma or maybe just from a lot of thinking that you came to that realization?

Yeah, I mean, I have actually— I'm not new to antibodies, but that's not my past. And so I've been connected to cancer in the cancer research space, which really preclinical models. I wouldn't even say it's cancer research. It's more like before you even get to the drug screening, trying to understand cells and pathways. I've been exposed at that for now several decades to the mission of trying to cure cancer, which is basically what everybody would hope for has happened, but we're very, very far. And so that mission has always kind of been lurking. How do I be part of that? How do I add some value to that mission? And so even the biosensor route, what we were doing where we were measuring other people's drugs, most of those companies had some connection to either oncology or autoimmunity. And so we learned more even in the early years of BioLumix, like 2019 on, We were learning by being boots on the ground for these people what it really looks like for the first stage of funneling drugs for oncology. And even that, we were surprised at how messy the search space is and how easy it is to find something that's not going to pan out either from day to day or going forward. We got a taste by being exposed to those companies on how you could galvanize a team around a mission like that. So I was already sort of on track, right? The company already had the idea that we were going to engineer proteins and the ultimate protein of highest value, highest impact is an antibody, I think, at this point. And then we got to see how these other companies did it, how they grew the company around this collective mission. So it's not just good enough to grow a company around technology. I don't think that will work. That technology has to solve a real-world problem, and that's basically an asset, right? That's like an antibody, a drug, or small molecule. And then the proof in the pudding is if it gets FDA approved. So it's a really long way to know if what you're working on is going to be successful. And so The way you galvanize the whole team in those companies we were exposed to and how we're doing it here is basically that mission is so big, you're trying to help so many people that it's okay that you don't know if it's going to work until 13 years from now or 10 years or whatever, wherever stage you're at. And so there's two answers to your question. It's like to build a company around a collective mission, the mission has to be big and cancer is a really big story. And then the other story is that we essentially are turning human cells into DNA encoded libraries. So we're just like Dell libraries for small molecules, we're Dell libraries for biologics, but we're using natural DNA, right? The cell's DNA where we've injected and added a gene. So it's the oldest school technology being reapplied basically. And because we're doing that, we need to be building an asset that's genetically encodable. So it's either a piece of DNA or something that DNA encodes. And that's why we're on the protein.

That makes a lot of sense. I think it's kind of if you're doing something interesting in biology, therapeutics is probably the most impactful thing you can do right now. And just kind of the reality of the market, maybe to shift gears. More towards bioluminescence technology and then, you know, kind of product plan. Maybe you could describe a little bit more of the technology. Uh, it's like, you know, genetically encoded libraries of proteins. Uh, how does it actually work? We have all— like, from my understanding, you have a system that kind of breaks out of the 96-well paradigm of screening, and you can be able to, like, batch together millions of cells that encodes its own unique antibody. You know, can you describe the workflow? And then could you describe, like, how do you— that early point you made, self-drugging cells, how do you actually measure that phenotype?

That seems to be the trickier part. So I'll start there. That seems complicated, but really FRET-based sensors are proteins that are inside cells, right? And they have two of these jellyfish proteins, a yellow one and a blue one. If the two fluorescent proteins are near each other, you can get FRET. If they're not near each other, you don't get FRET. So it's basically, you can use a microscope and shine light on cells and make a measurement, that FRET measurement. And these proteins, these genetically encoded proteins, when they get phosphorylated, they change shape. And so FRET is related to a biochemical event inside that cell. So with light, you can continuously read out how these proteins are being phosphorylated and dephosphorylated. And so that's how you can not destroy the cell and constantly measure some pathway. And the trick is, how do you get a pathway to phosphorylate that protein? You mutate it until you find a peptide sequence that the kinase likes. Or the phosphatase doesn't bind. It's really simple. So then, so we take those types of assays. So that's one type of assay where we can measure things. The other type is you take a target, put a fluorescent molecule on it, and you can see where it goes in the cell. It could be at the membrane, or it could be in the nucleus, or it could be in the ER. And so there's a number of these tools that make cells fluorescent, and you can measure things. To make a self-drugging cell, like if you took an antibody and you just put that antibody's DNA that codes it into a cell, that cell will just secrete the antibody into the abyss of the media around it and then drug all the cells around it when it accumulates. And you won't even be able to tell which cell makes that drug. So that's the challenge we had to get started with. So we figured out how to make an antibody with a tether, a tether that kind of anchored it back down to a cell. It's able to be secreted, so it's not like a ball and chain model. It's more like Velcro, right? So the antibody gets secreted and it has a hard time diffusing away from the cell because it's always pulled back by electrostatic interactions. And so cells that make a variant experience a variant just because we added this one 20-amino acid addition to an antibody's end, its C-terminus. So that was like a eureka moment where we were like, okay, those cells are the ones that make the antibody, those cells are the ones that drug it. And then we iterate on the DNA sequence that encodes that antibody to try all new wacky crazy ways to build antibodies. And so these days, what we're only doing is going after antibody degraders, antibodies that move targets from the surface of a cell to the lysosome of a cell, right? So we fluorescently tag that target from, and we can measure where it is in every cell. And so most cells, for example, is like EGFR, That target's always on the surface. So you take a picture of all those cells, it's fluorescent, it's pretty awesome looking. It's like you see the membrane of every cell and you start programming all these antibody constructs in. And then one day in like 2022, we saw a rare colony where we had just started picking up random ideas of how to build antibodies and that one had dragged a target EGFR to the lysosome. And we said, aha. And we followed through on that antibody. We got the sequence of it. And we basically saw a potential of membrane-binding antibodies. So we, using that sequence, we figured out if an antibody binds a target and an antibody binds a membrane, it's a degrader. So we invented an entire modality without any models. We just searched for when random fusions of different proteins and peptides to antibodies gave the effect of moving a target from the top, from the plasma membrane to the lysosome. Totally interesting. Pot full of ideas. The only thing that entered that pot were, were things that we could think of. And so it was just basically like a matter of time until we found a solution and then started to figure out how it worked.

Was this before, or maybe around the same time the LyTag stuff was coming out?

AbTag, —exactly the same time. And that inspired us big time. So, so it's very different. So LI-TACS, AB-TACS came a little bit after. So it's basically Jim Wells and Carolyn Bertozzi's lab. And so, so they, they're really pioneering. So Carolyn Bertozzi's Lycia and LI-TACS are essentially a conjugated antibody. So you take a small molecule, 6-mannose phosphate, you put it on the, on the antibody,, and there's a receptor for that molecule. So you start taking two activities and putting them together into one protein. And that's what we're doing, but we don't have a receptor. That's one of our differentiators. So the ABTAC is basically, there's many versions of it. It's a bispecific idea where you take two different binding events, you put them on one molecule. And one of those events is something that rapidly cycles in a cell from the plasma membrane to the lysosome or some other compartment. And the other one's the thing you want to drag there. And so when those stories started emerging was when we were like, aha, we have an entirely different way. And we haven't even published on this yet. I think that's coming in 2024. So yeah, I'm very excited because I'm like pushing everybody. I actually started BioLumix saying we'll never publish again. But now I'm like, we're publishing. And so like, this isn't the time. Of wild, crazy antibody ideas. And it started, I would say, Lycia and Carolyn Bertozzi, that kind of kicked it off, this AbPak. And so EpiBiologics is one of the companies too. And people are basically saying, IgGs basically don't do the job we're asking. They're too analog. They're the idea that you just bind and block. And like toxins, so I'll go way back to the DARPA project where we studied toxins. Toxins do 5 to 10 bits. So like cholera toxin is like an antibody degrader. It binds the surface, it binds the lipid, it hijacks its pathway. That was actually one of the inspirations for our modality, which we call MEK antibodies, membrane-engaging chimeras. And so cholera toxin binds the lipid on the membrane, it traffics to the ER, it gets cut up, it starts doing another thing, it opens up, it enters the cytosol, and so forth. So, basically, inside that molecule is like 10 events programmed in, 10 different binding events and sequences. It's not binding and blocking something. And you don't want to get cholera toxin in your body, right? It's a very powerful outcome. So, if we want very powerful outcomes, we should get away from these monovalent ideas of this magic bullet dumb molecule that just binds something and we kind of just like pray that it's going to do this magical thing that evolution says really needs many valencies, many different binding events to achieve.

So on this point, I like biologics. I totally agree with you. Last maybe 30 years, 37 years, antibodies have been kind of just one target and You know, it's worked out pretty well, I think, uh, for patients, but there's a lot of room for improvement to add new functions, either multi, uh, bispecifics, trispecifics, multifunctional conditional biologics. And, you know, I think you're seeing a lot of other companies and labs do that work, like Virtuosi and Wells, but it's not— it's like, from a rational perspective, a lot of it just from like, oh, we want to add conditionality to our biologic. We're going to add this domain. We're going to do this FC modification. You know what I mean? Like, we have a hypothesis and do it. Well, biolomics has just used evolution to invent. Yeah, yeah, it's, yeah, it's a whole different— yeah, sorry.

Oh, I would just say that like, um, the difference there is synthetic domains. So, so like, um, when So we're trying to build these peptides that bind the membrane selectively. So you need to sort of sort through, like, too sticky, not sticky enough, too charged, not charged enough, doesn't fold well. When you go into the synthetic world and try to make de novo proteins, you find out how rare an active protein really is and how hard it is to build something de novo. When you iterate on a protein, so, like, Antibodies, bispecifics are a good example. You take a scaffold and you mutate in a specific spot on the scaffold. You don't create the scaffold all at once. And so I think directed evolution is the only way to make synthetic domains fully from scratch. But you need to have huge throughput power. You need to be able to test lots of ideas. And you need to use AI to analyze the sequences and basically explore beyond your wet lab testing into the full chemical space. And so that's basically why one reason we went there was we got a eureka moment and saw something work. And it could have been one day we never— it could have been we never saw that. And then maybe Philomix wouldn't be here today or maybe be doing something very different. But this is an example of data-driven pivots to where you're seeing your tools excel. And so we saw our tools excel at showing us degraders, and now we're exploiting our— we could be making regular antibodies that just bind and block if we wanted to, but that would be doing the same old thing with new technology, and we're trying to really push the bounds. And so here we are today, we're basically doing this whole next evolution of BioLumix where we're taking all these sequences. So now we have 570 of these peptide sequences that can make degraders. They're all very— some of them are okay, some of them are great, some of them seem more manufacturable, some not so much. And so the chemical space is almost a million-fold beyond what we've tested, just even with all the constraints in for the scaffold. And so we need machine learning. And so we're doing this whole new stage of BioLimx where we're taking the sequence data and we're basically doing large language models to try to predict the activity from what we have so far. So sort of like most platform companies, I think you probably hear about these days, once you find your training data, you become very ML-centric. So sort of first is testing, can you generate the data? Then it's going to be ML, and then you're going to sort of leverage an iterative feedback loop to find a drug that you could never have found by random walk.

That makes sense. I think that answers my first question. The second question I had was the importance of a starting point for your models and for you to discover. We had a conversation last year with Bruce Whitman, who led machine learning at Francis Arnold's lab. He works at Microsoft now. But he did a lot of ML-directed evolution work there. And the number one thing he learned from that experience is starting points matter more than anything. And so for biologics, like maybe for the, the greater discoverer, how much do you account for the starting point of the scaffold you screen for generating that lead? And looking forward, how are you trying to diversify not only the stuff you— the— not only like the, the data you generate, but like the, the starting points?

For what you screen? Yeah. So starting points have to be on some mixture of rational design and random chance. Yeah. And so the beauty of DNA libraries is that you can constrain some spots and go wild in other spots. And so that's how you can accidentally find things that you didn't fully design. But you knew there were some first principles behind it. We knew membrane binders need to have a little bit of hydrophobicity, but if they have too much, they crash out of solution and then you have made an aggregate of protein trash. And so we knew these principles that are the extremes, and it's right in the gray area that you don't know how to find the perfect one. You need random chance to find your starting point. So when we were at that time, we were kind of coming up with all different ways to bind all different things on the membrane of cells. Most of them were protein ideas. And the ones that ended up being these small peptides that had a little, just the right combo of like they could be made, they could secrete, and then they were, they would bind the membrane. They won by fortuitous chance. They were in the library. So we had some rational design, but there was a lot of them in that library that basically followed the exact same first principles that didn't shake out. And so, so, so starting point needs to be blurry is one of my, like, like where, how you find starting point needs to be blurry. So you find something that's robust. So that when I saw that first colony where we had moved EGFR to the, to the, it was actually the late endosome that was glowing, because when it gets to the, to the cellular trash dump, the lysosome, It doesn't fluoresce as much or at all anymore. So we basically saw dots instead of this membrane. And I knew that, like, from a mile away, I've seen enough cells that that is a strong protein. Whatever that did was very robust, and it took down an overexpressed target and it moved it all 100% to this organelle. And so we followed that because it was so strikingly different. Right next to that colony was probably the same story, the same principles that we had baked in, but it wasn't the right combination of amino acids. So we didn't know how to solve the problem. But when we saw that and we saw that some of the losers, the ones that didn't have the degradation power, were very similar, we also knew that there's an evolutionary landscape. It's not just biophysical drift. Where there's very close relatives that don't have the same effect. And so basically, if you ever want to find out if anything has specific rules, just shuffle all the positions in the protein world. And if you have the activity, it's not really like from folding or structure or anything like that. It's biophysically driven. Ours is a combination of that— spoiler alert— but it was really like, so that starting point had to be so convincing that I would bet the company's future on it.

It's pretty terrifying, but it worked out. So on that point, do those same principles apply as you scale up your platform? Because by Illumix now you have degraders, but you're also doing ER-specific activity— no, organelle-specific activity— ER, Golgi, lysosome. Are those same initial principles that got this platform bootstrapped Are those the guiding principles that keep it going, or do you have to add new capabilities?

Yeah, so the reason that that's there, so that's like on our website, that we, we're very focused on the ones that go to the lysosome. Yeah. But we watch for other ones because they essentially, if you could delete a target from the surface, that's probably all you need to have therapeutic value. Because its action is at the surface. We are going down a road, the company is evolving where we're degraders and we think the best spot for degraders is in the ADC space. That's why the lysosome is the best endpoint. But we, we're collecting, we're amassing a data set of peptides that do things, that traffic antibodies from one location to the other. And that might be valuable for a future pivot or might be valuable for a partner if we partner with people. But it also can be valuable for our machine learning. So all of those peptides, no matter which organelle they bring to, bring a target to, they call it endocytosis together, but they fate to different locations. And so you can really start to tease out through just this evolution process how synthetic proteins can become like mail addresses for an antibody. And so rather than ignoring that fully, we just kind of store them in piles and we really focus on the lysosome ones, but we see ones that get trapped in the Golgi. We see other organelles too. And so, you know, that's sort of there for a call for anybody that thinks a real therapeutic solution would be one of those because we do have them accumulating in the background.

Cool. And on this point of generating data to make a therapeutic, but also generating data maybe to do partnerships, what kind of informs your business model and what inspiration do you draw from other companies you like? Is there certain problems that motivate you? I think cancer was one of them, but how do you go about developing a platform towards a drug? But also scale it up so maybe you can do deals with other people if it's appropriate, or if it's something that's not a non-focus area.

Yeah, so like, don't get me wrong, I want, I want the company to be— to pay out everybody and grow and be amazing, but I want to solve a big problem with what we're working on. And the, the, like, that, that mission thing is, is what will keep us all together. The cash probably won't. And so it's not just that cancer is a big story and that there's lots of people dying and probably me. It'll afflict us all eventually if it's not already. But I see something really big. So as we went to the degrader space, we started to survey where degrader is going to be good.. And we got to this point where we said, aha, ADCs are going to be the best spot for a degrader because they have— so ADCs are an antibody, they already have a conjugate on them, the conjugate's some chemotherapeutic or some terrible toxin, but it only comes off that antibody and becomes active if the whole thing fates to the lysosome. But most drug targets, they're, they're basically on the surface doing lots of things. And that's why cancer selects them. And by doing that, the half-life's been driven down, driven, excuse me, way up. So they just stay put. So when antibodies bind that, antibody drug conjugates, they don't really fate to the lysosome. So most ADC targets, I basically think, are terrible. I'm going to make a prediction. All antibody degrader companies are probably going to become ADC companies one day because that whole This entire modality has a super strength. When an ADC kills the cell that took it in, so that molecule goes into the cell, goes to the lysosome, releases the toxin, toxin does what it's supposed to do, it's so bad it kills the cell, then it kills the neighbor cells. That bystander effect is one unique thing that does not happen in any other therapeutic strategy. And so, so when we're wrong about a target, What ends up happening is a drug is made, it goes into a cancer patient. The tumor is an ensemble of different types of cells and the drug kills off, hopefully kills off all the cells with the target. And we find out, oh, there was a bunch of other cells and those other cells don't have the target and that drug's useless for them. And then the cancer's back. And this is kind of like the age-old story of a resistant population taking over, being worse than original, and then you die of that. ADCs have this superpower. They can kill a cell next to the cell with the target. As long as it's a well-mixed tumor, you can kill everybody with the wrong target, as long as some of the cells have the right target. That superpower is the promise of ADCs.

Now, the shortcoming— To clarify the bystander effect, is that due to apoptosis or—

Most of those small molecules were therapeutics. Yeah, exactly. So they'll kill cells if they see them, right? If they get into them. And when that cell dies that had that toxin in it, it didn't destroy the toxin. The toxin just goes to the next cell and it goes to the next cell. That's one of ADC's downsides is the toxins are so toxic, the toxins can end up somewhere after they've done their job. And so you can get side effects, go blind, things like that.. And so if you can change the turnover, so like antibody degrader story, if now an antibody degrader has a cargo on it, a toxin, it'll always go to the lysosome. So less binding events equals more cell death. You could use less toxic cargo because you have a continuous highway into that cell. So less molecules kill the cells. That means you could use the same amount of molecules with less toxic cargo, or you can use less molecules with the same toxic cargo. It opens the door for lots of things. It opens the doors to redesign ADCs. It opens the doors to go after targets that sit there and ADCs don't do anything to. So they just like coat the surface of a cell. The toxin never gets the lysosome. Nothing happens. And people say, great, our program didn't work. That toxin that target's no good. And if you have an antibody degrader in that antibody spot instead of just a binder, then it doesn't matter that biology has told that target to sit at the membrane. It overrides that and it just drags it to the lysosome.

So like I said, I think every antibody degrader company one day will become an ADC company. So, you know, just by the nature of being able to recycle its cargo, which kind of creates like a major amplification. And, you know, I think that kind of solves— a lot of ADCs have this toxicity issue, and conventional wisdom is to solve the linker problem, but maybe just taking a different approach of just a degrader antibody. I don't think anybody's really thinking about that except you guys.

And so, yeah, it's so new. That, um, you know, there's only 3 modalities and we haven't even published ours. So it's like, I, I wonder if there's another us somewhere listening to this podcast and they're like, we have another solution that's a degrader.

Um, on this point of like developing something kind of a breakthrough technology, how did— how do you go about thinking about cancer and maybe developing an asset towards clinical trials? 'Cause the last 5 years in oncology, it's become a little more broad-based for each asset where you can take a basket trial approach initially and then get the details figured out post-phase 1. Or do you have like a specific hypothesis for like a particular disease? At least for oncology, that's what's changed the last 5 years from at least the pharma companies I've worked with. Sometimes in cancer, you don't have to have it all figured out. You just have to get acid that's safe and ready to go in humans. Is that what you have, or are you a little farther in terms of like—

Well, we're still quite preclinical. So the way we approach this problem is that we have many, many programs. Some of them are safe, and some of them are exactly what you're getting at. They're a wild, crazy idea to leverage an entirely new thing that we think might actually kill cancer. In a broad way that it would scavenge the cancer regardless of targets or generally of targets. And so we have that program. I mean, I'll describe that one a little bit because I like that one the best and I think we're probably the only people positioned to even try. But this protein CD59, it coats every cell in your body and it basically is the self, the universal self system. Signal, and your complement won't poke holes in our cells because of that protein. So if you scan all the single-cell data, every cell has CD59. It's ubiquitous. So if you could degrade that specifically in a certain cell type, like a tumor cell, so if you couple to a tumor-specific antigen binding event, and you basically, when a cell has HER2, you degrade CD59,, you can really get a complement wildfire attack of the tumor. And we're sort of suspicious that that should actually also create a tumor microenvironment that would be like this bystander effect of ADCs. So we have a pure degrader story in the mix. There's no cargo, there's no nothing. So that's like our wild pie in the sky crazy program. And animals will tell us if it's better than all the other ideas. And through these other programs, we're basically saying known targets are good targets. The problem is the cargo never gets to the lysosome. So that story I told before, better ADC. And so you run all these horse races and animals basically stratify what you're going to bring to the clinic. And what you really want is that whole story back where I said the most robust thing is where you're like, aha, I'll bet the company's future on this. And so, so, you know, going forward, these are very diverse ideas. They all test where, where degraders will be the best. Will they be the best at new targets that we could never go after? And that's that CD59 story. Will they be the best at improving things that have super clinical validation but don't have super efficacy? And it's just what they do is they improve efficacy. Or they excel at making safer ADCs where you can use less toxic cargo and all the side effects disappear. And so each one of our programs is a version of those stories. Animal data will guide us. So I can't even tell you where we're headed.

I can tell you what's on the list. So on this point of just being informed by the data, what's your long-term view on— or what's your— not long— just a general view on startups, especially scientists starting companies. What do you think the prerequisites are? How do you build out that capability of truly letting the data inform your decision-making rather than having some sort of iterative approach that the academia promotes or incentivizes? What's your view on like what it takes for a scientist to start their company?

Yeah, constant rebuilding. So, so you are not you as you knew it last year. Every year you're an entrepreneur. And so, so as a scientist, you're, you're basically like— I speak for myself, I felt like I was in a cave for 7 years, and then I emerged and I had to rebuild how I'd interact with people And so I had to start paring down the walls, but really everybody was tearing things off of me because in order for us to get cash, in order for us to solve problems, in order for us to work together, all my hangups had to go away as do other people on the teams. And so you have to be risk adverse at the beginning where you're like, I'm willing to do something that might fail. I'm willing to quit my job over it. And then you're like, oh, my personality's in the way of being a leader. I need to change my personality. So you change that. And then you started out being risk adverse. So then you realize, oh, you actually do have to weigh risk because that's how you guide success. So you can't go wild, wild west and just start making random shots into the darkness. You have to be very strategic because you have limited resources. So then you start to learn like a true entrepreneur about resource management and how the balance of risk-adverse and risk-prone right in the middle is where you got to go. You got to constantly readjust where the data is pointing you and you got to sort of listen to everybody where normally when I was in the cave, I was just shouting at everybody. But you got to listen to everybody because What you're building needs a market to buy it. And so you start to learn what a market is. And so you truly become not even a scientist anymore. You're like a technical founder basically has to get rid of the technical part on some big level. It's there, but you have to totally change the way you think that you need to talk to people, learn from them. Bring them on the mission, and then together you're stronger than you by yourself. And so you have to sort of see it as like the, you know, like, you know, people like the PBS world would be like, like everybody, everybody is a team and teams are stronger, right? But I don't think people believe that until they see it work and until you need it to work. And so like in science, you can silo and and you can be very successful. You can be a professor and be siloed. But as an entrepreneur, I don't think you can do it. I can't even think of somebody that's done it. So you basically have to be okay that you don't know what the future holds. That's the risk adverse part. You have to learn from the people to know what the market wants, what is the possible for you. It's not just from what the data can do. It's not just from what the technology can do. It's not just what you can build. Someone has to buy it. And so you have to see these resources that you don't have, how you're going to get them, and how you're going to apply them. And so it brings me back to this idea that it's like writing a game or a story. And, but in this case, the world is changing and you're trying to bounce the ideas off and see to see what things fit. And so you can't just write the book and be like, let's execute. It doesn't work. The world is different in chapter 3 than you thought when you get there. And so, I mean, my advice to entrepreneurs is you just basically have to be okay that you can see only as far as you can see, and then you do your best to make sure that all the pieces are in place.

And adjusting every time the world changes. Well, that's a great capstone to the conversation. I think, uh, I'm excited to like talk 2 years from now-ish, and I think this year is going to be a big year for biologics. Um, but Doug, I really appreciate doing this, and, uh, you know, thanks for the time. And I think a lot of people are gonna get a lot of value about this, and I think all of us are going to be really excited for BioLomics this year as you make some announcements about these kind of programs and go into more details.

Well, thank you, Josh. This has been an absolute pleasure, and I really enjoyed it.