Rare Compute, Fighting Rare Disease with AI & Web3

GM, GM. Welcome to what's going to be another great DevNTell. So if you didn't know, DevNTell is a 30-minute podcast held every Friday, allowing founders, hackers, and anyone in between a stage to showcase their product. And today, I'm ecstatic to welcome back on the show Stanley Bishop, who is the co-founder of Rare Compute. Rare Compute is a decentralized biotech platform that aims to revolutionize rare disease research by leveraging blockchain technology and computing credits. If you stick around for today's episode, you'll see Stanley give us an overview of Rare Compute, the story behind its origin, and their latest research in the fight against rare disease. All right, let's get into it. But before that, a word from our sponsor.

Today's episode of DevNTell is brought to you by our friends at Arbitrum. Arbitrum is a technology suite designed to scale Ethereum. You can use Arbitrum chains to do all the things you do on Ethereum, but your transactions will be cheaper and faster. With over 3.8 million ETH saved, 1.69 million active wallets, and over 1.7 billion total transactions, why would you want to settle for anything less? Get started today by exploring Arbitrum's comprehensive docs and tutorials linked in the description below to help you get started launching your next dApp today. Now on to today's show.

GM, GM. Welcome back on the show, Stanley. Happy to have you here again, man.

Oh, man. I'm so happy to be back. How are you doing?

I'm doing well, doing well. Excited to learn about all the great work you guys are doing at Rare Compute. As we know, the fight against rare disease has been all too hard to fight, so hopefully there is some great progress you and your colleagues have made in the area. But I guess before we get into all that good stuff, if folks didn't catch you while you were here the last time around showcasing Lilypad, did you want to give a brief introduction about yourself?

Yeah, absolutely. My name's Stanley Bishop, and seriously, man, so happy to be here. Always, like, even when we do a one-on-one call, I always have the best time, so I think this will be a lot of fun. And yeah, my name's Stanley Bishop. I'm a mathematician and a computer scientist. I unfortunately think I have to say also a patient. I have a kind of rare/complex disease, and I'll tell you, not having the best day today, so do forgive me if I, you know, stumble or fumble as I'm speaking. But yeah, I'm a person who, you know, had some roles in leadership in machine learning in kind of the early days of using GPUs to accelerate machine learning. And it, I think, gave me an interesting and early insight into what was going to happen and what's been happening with AI, and particularly how AI can be a positive part of an existing system, but then also how AI can disrupt existing systems, often in ways you weren't expecting. So, you know, that very much, given my experience as a patient, brought me to the idea of how can technologists bring some of our tools to that world, integrate them positively with healthcare providers as partners, and yeah, hopefully improve the outcomes and experience for patients. You know, that's kind of what I've decided to be dedicated to.

Amazing, man. And I guess you have quite the background in AI. You used to work at Google, I believe, in the machine learning vertical, and you also founded a couple startups of your own. I guess did you want to kind of lean into that a bit? If there was any inspiration perhaps from your rare disease going into those fields or whatnot?

Yeah, no, super happy to, and you know, I think in particular an experience I had building computational linguistics technology for Google was very formative. I was the principal scientist working on a platform called Falcon. It was a language translation platform, and this was like 15 years ago, so in the days of like, you know, you wouldn't want to use Google Translate for anything real kind of thing, right? So at that time, you know, we were building technologies to accelerate translation, to support translation and localization, also to get the right jobs to the right translators and localizers. So in that sense, you could think of the system as Uber for language translation. And kind of a fun like little statistic, I think at the time I was working, our group was responsible for I think something like 20 to 30 percent of the world's total daily translated language. So yeah, really was so fun as someone with a background and an interest in computational linguistics to kind of have that bird's-eye or I guess you'd say that Falcon's-eye view of just kind of how the flow of language works. It was really a special thing to be involved in. But yeah, I'll tell you, it was actually a population of about 75,000 translators all over the world that the Falcon system, you know, flew between. And so really, like I said, a really interesting thing to see when you have an existing system with so many humans carrying out so many actions every day, putting out so much effort, you know, what happens when you add these machine learning pieces to the puzzle? And can you always predict what they do? And yeah, turns out you can't. But yeah, there really is a kind of a science, you might say, to how you kind of implement those systems. Sometimes this is called cybernetics. You know, we often hear about cybernetics and we think about like putting robot gear on your body, but it's also, it's like being in a system that is, you know, containing human and non-human intelligences. And let me tell you, I think we're seeing we're all now part of a cybernetic system, right, Narb?

Yeah, yeah, 100%. We're leading that way. And I mean, it's just going to get even more, like, proficient, like us being one with all the AI and whatnot that's getting developed. Like, I think we're moving at a really rapid rate. I'm sure you've seen it. I know I've seen it around.

Man, we'll have to talk later about Claude Code, which is such a takeoff point. But no, keep up with even, you know, my little slice of one little slice, you know? But yeah, interesting seeing it all play out. And yeah, it was like I saw some really interesting things, like kind of just how humans react to being involved in machine learning, how people react emotionally to the idea of, you know, interacting with AI. And yeah, became really interested in sort of exploring the frontier of how those technologies would be applied. So yeah, after leaving the project at Google, I founded a really interesting alternative startup studio in Venice Beach that was called Space Post Labs. And our theme was impact venture. So we had this idea that sort of like being in venture isn't just a really exciting and often lucrative position, but it's a chance to be, yeah, at the frontier of exploring how new technologies are going to interact with communities, human experience, human health. And so it really I think should be a responsibility to people in venture to kind of like, you know, when it's possible, like do good and do well. And then I'll tell you a secret, in the venture world, that's like oxygen for people. You know, I think people are so used to a very different way of doing things that, you know, you can kind of set that dual motive and accomplish some really, you know, positive impact milestones, but also, you know, people will react to you differently. Like, there's advantage to be found in that part of the community. So yeah, was really so fun, and we operated for about 10 years. We were part of a really special art patronage network. We were like their kind of technical provider called the Budman Studios. And then yeah, about a year or two ago, Jim Budman, the founder, retired, and so kind of happened to be at that time, you know, had this really interesting experience at something called the Stanford Rare Disease AI Hackathon, which, you know, I think that would be such a cool thing to, you know, start a rare disease story, the Rare Compute story with, because yeah, man, that was literally what it all grew from.

Oh, yeah, yeah. Well, we'll definitely get into it. And I guess from your venture days, was there any particular type of company, company profile you were looking for when you were investing? It seems that you were looking for a very particular type of company there.

Yeah, would totally say that it was companies, but I always thought of it as founders. Like, I was a very founder-centered builder. Founders I believed in and founders I thought I could help, you know? Because I'm not perfect, I only have the cards that I have. So I think being realistic and judicious about where you can add value is something anyone in venture should do. But but yeah, I guess I would say like definitely machine learning, you know, like frontier data technologies were a big piece of it. But I was particularly passionate about finding founders who like I guess you'd say normally wouldn't be hearing about that stuff or wouldn't be exposed to those things. So you know, I think actually if I really went down a list, like we, you know, I think we built the architecture for the world's most successful digital art gallery chain. You know, we also helped develop a bunch of like neurological treatment courses for some diseases. So you know, really everything from entertainment to STEM and everything in between. And you know, branding-wise I think that's what we called ourselves, a STEAM incubator. And so, you know, if anything, the theme was really bringing diverse voices to the table and kind of seeing, you know, what those voices that might not otherwise have a chance to talk get to say.

Amazing. That's awesome, man. And like you said, you found yourself at a Stanford hackathon. I'm really curious. I don't know the origin story of Rare Compute, so let's get into that, if you don't mind.

Yeah, well, also ties into my broader origin story as like Stanley the DeSci Guy. Yeah, 100%. You know, it's, what would it be? It's like wish.com Bill Nye character I try to do, but I I do have fun with it, so that's a plus. But yeah, I used to be I guess in addition to like the work I did in venture, I am a researcher and have continued my life as a researcher and I work in, you know, hyperscale computing, specifically for kind of post-training and application engineering in AI. So yeah, I'm really interested in like how do you properly set up and operate, you know, thousand, ten thousand plus clusters and you know get them working for scientific computing purposes. So yeah, I have always kind of like worked on projects as a research collaborator or contributor. Particular project got a shoutout because it was a big part of me entering this world was DeepChem. Really special project that grew out of the Pande labs at Stanford, but it's I think probably the most high-quality and, you know, performant open source machine learning technology for medicine discovery and molecular engineering. And yeah, like kind of got involved in sort of medicine discovery and bioinformatics as like someone doing the computers at DeepChem. But then yeah, I have a background as a physicist and love to learn and I think more than anything I really love working with experts and sort of helping them communicate, collaborate, you know, activate large computer systems, you know, work together to get them doing useful work.

So yeah, my work at DeepChem grew into work with a lot of different places because it sort of happened that there was a big realization and a rush to kind of get access to these machine learning technologies, like actually particularly in the wake of AlphaFold, which, you know, might not need an introduction but happy to talk about it anyway, although you'll you'll have to cut me off because I can go on forever about AlphaFold. But but yeah, so it was like sort of had this career and practice going on for quite a while, started to get interested in DeSci and passionate about DeSci, and then I think it was like three or four years ago I was like serving as research architect for the Snyder Labs at Stanford Medicine, that's the Stanford Genetics Department, and had like really a cool role with a particular group that looked at frontier technology application in rare genetic disease and oncology. So we were often working with patients who'd completely exhausted traditional course of care or who had diseases where there was no course of care. And, you know, in many cases, like with the rush of AI models, with, you know, many new incredible technologies, like there are, you know, things where you can really change the outcome for a patient if you kind of build them fast enough, teach the doctors how to use them fast enough. So yeah, it was was sort of a kind of triage technology architectural and really a lot of fun, you know, just day-to-day building cool stuff, having really I hope positive impact on patient outcomes. But yeah, a kind of funny story starts with an external hard drive. We had a particular patient who had a condition called EDS, and this patient was having the classical rare disease story. Struggling to find doctors who knew anything about his disease, struggling to interact with administrative and insurance layers of our medical system, and even things like, you know, his medical records, yeah, like literally filled a 4 terabyte hard drive. I think it was 3.8 terabytes of medical records, different formats, PDFs, like just a mess, you know? And the guy was going to doctor's appointments and he would spend the whole time, you know, when he should be interacting with his doctor and getting feedback, like trying to find that one form. So yeah, this was just in the moment when RAG systems were available, like retrieval-augmented generation technologies. This was the first moment when people were connecting multiple agents to serve purposes. So we kind of got started like just building a RAG system for this gentleman. But yeah, we realized like as soon as this thing was built and everyone was excited about it, you know, I think the doctor we built it for said, you know, she didn't have a migraine for the first time, right, because her and the patient were able to just kind of find what they needed. Anyway, we ended up getting a lot of demand on the side of other physicians, like curiosity, how do I use something like this? What does this thing do? A lot of it was positive curiosity. I'll tell you there, Narb, we we did have one funny thing where an oncologist discovered hallucination of LLMs, sort of in a funny way. This scientist asked for a bunch of papers on the interaction between viruses and oncology. And a paper popped up that she had been writing with a colleague, the exact same title, but the colleague's name and not her name. And this was actually a total coincidence, I mean not like a coincidence, right, because when models hallucinate, they hallucinate things that sound dangerously reasonable. So the model actually, like based on the previous work of this scientist, predicted she might write that paper and put her name on it. And then so it was just anyway, I think there was an angry phone call that was probably very confusing on both sides. But anyway, that's all to say like I think, you know, many of us are familiar with hallucinations, familiar with some of the complexities of these systems. But yeah, this was really the first moment when the medical system was thinking about this stuff.

And then another really cool thing happened when we started telling people in the technology community about like having done this service for a patient. Everyone was like, is there anything we could get involved in? We really had like just when you shared the story, everyone literally was like, how can I help, you know? So we decided to see if we could sort of scale that. And so we, you know, created some structure. I designed a research methodology where, you know, we would be kind of standardizing the tasks, the datasets, and the models, and then providing validation, you know, so providing medical leaders to review the systems. Validation and model post-training is kind of my whole field. And so one of the things we built, Narb, I think you would have thought was particularly cool, was almost like an arcade where our doctors could play the models in the sense that, you know, these models that the teams were building, they could log in, talk to them, and then correct them. And bro, let me tell you, if there is anything doctors really like, it's correcting stuff, so we actually got some really good user engagement there. But but yeah, then we actually used a modified statistical system. Have you ever seen Chatbot Arena?

No, I've never seen it, no.

Oh, you gotta check it out. It's so much fun and it's the really set of benchmarks that I trust most for model performance because it's totally human validation. So you go on there and just like I was saying our doctors did, you say, I'm asking this question, here's answer for Model 1, here's answer for Model 2. You know, it's Tinder, it's like you swiped left or right, which model do you like? And if you've ever played like chess or maybe StarCraft is also a place where there's a numerical rating, and basically each time you win a match, you gain some points, each time you lose a match, you lose some points, and then the number of points you gain or lose is proportional to the difference between you and the person you beat or were defeated by. So this is actually a system called Elo, E-L-O, and it's yeah, really beautiful statistical system for providing merit-based tracking of performance between things that can undergo heads-up competition. And yeah, yeah, so it's so cool and was really interesting to kind of see how doctors interacted with it, see what the behavior was of the models within the system. And yeah, we did kind of find that like the medical use case was very different. For example, medical models can give answers that are disqualifying, right? Like if they say something that is potentially dangerous to a patient, then that model has to actually be removed from the system until it's triaged.

Similarly, and very similar actually I would say to the human practice of medicine, kind of alluding to some of the cybernetic concepts here, doctors are very specific and rigorous and when they change their procedures and practices, it's very rigorous, it's a rigorous process. They do very deep and systematic review, and so on the model side that looks like kind of making the system a little stickier. So if a model is starting to win, we actually tweak the system so it climbs the ladder a little more slowly, right? So anyway, this was just like the kind of idea of this hackathon, a decentralized hackathon where all of the people with knowledge and wisdom about rare disease all over the world could kind of contribute their their spice to this stew. And oh my god, Narb, it was really beautiful, the response was like insane. You know, I think in the end we had over 250 like doctors involved in different ways. We had maybe 500 volunteers and created over 30 teams from over 20 countries. And man, was quite a bit of unexpected activity, 100%, but really was a beautiful thing to see. And then just all on all sides of of this situation, you know, like rare disease advocates and patients who kind of struggling to find support, doctors who who need assistance processing all the data of rare disease, and then also like technology companies and engineers, like who you know are all human and all have families, but then you know, quite frankly, I think all of us are pretty concerned about what AI is going to do.

So it's like really a beautiful thing, you know, it's I think really easy to say that like, you know, the world is in a place of, you know, people not caring and, you know, things not working. But then you know, maybe it's just we need to think about how to create those opportunities and get them in front of people because like yeah, it was so incredible. Like again, like we said, like we got a jump for the patient and everyone was just like how high, you know?

Yeah, that's awesome, man. And like you said, you found yourself at a Stanford hackathon. I guess from all of that, it sounds like that was kind of the major springboard for bringing together a group of people to start Rare Compute, if I'm not mistaken, right?

Yeah, well, it's just yeah, get carried away talking about how how fun it was and how how much excitement and positivity it created. And one last note, Narb, I got to tell you just one last like cherry before I do turn to the dark side. I'm just kidding. But do you know the OpenAI co-founder, Greg Brockman?

I've heard of him, yes.

Oh, I mean like as a machine learning scientist, one of my biggest heroes as a leader, you know, not just a very brilliant scientist but, you know, machine learning scientists are smart people but complex people and like being able to lead and build organizations, that's like such a special guy. And anyway, we found out like just a week or two before our demo day, which was at the GitHub headquarters, that Mr. Brockman's wife had EDS. Actually, she had just been recently diagnosed with a particular variety of EDS called hEDS, and she had had the exact same set of experiences as every other EDS patient I've ever worked with. Chronic issues with health throughout her life, no doctors who ever really gave any indication of what it could be, a lot of doctors who maybe didn't take it seriously, and then eventually she found a caregiver who had seen EDS before and worked with EDS patients and immediately said, you have EDS, you need to get checked out for EDS. And then that led to knowledge of the diagnosis, which is for so many patients, like even sometimes when you get a bad diagnosis, when you've been working on it for four years, it's one of the best days of your life. You know, that's a funny thing we hear from patients. But but yeah, also, you know, EDS is a disease where if you have knowledge of your condition, you can really drastically improve your quality of life. And so it just goes to show like we all are human, we all have the sort of shared heritage of this biology with all its goods and all its bads. And yeah, was was just a really really cool special experience.

The thing I would say though, and can kind of start it with a joke, if you think doctors need to have good bedside manner, machine learning scientists who work with doctors have to have like really good bedside manner. And yeah, no, just kind of half kidding, but but half serious that there were some real human communication and coordination problems. The world of medicine organizes itself very differently from the world of tech, there's you know, different values, there's different understanding. There's also a thing where a lot of the work that is done on the technology side is kind of invisible on the medical side and vice versa, quite frankly. So it it really is something we realized there is some extra intentional and important work to do in like kind of building cultural bridges and building, you know, shared knowledge, like shared motivation, shared mission. And yeah, like we kind of just came away from it having seen, you know, playing this role, talking to engineers about rare disease, talking to rare disease scientists about machine learning, talking to doctors about the idea that like we're here to help, you know, we're not looking to replace doctors, like we know doctors are overworked, we want that to stop, you know?

So it yeah, just kind of like came to feel to me and a number of other contributors that there was just an enormous amount of good to do. And then to bring like one more kind of punchline that I think segues really nice into other things we want to talk about, it is a very I would almost say easily solvable problem. It is a problem of coordination and resource allocation. Right now there are many diseases that with the kind of current state of in silico tools we could make serious attempts at curing. And there is enough compute at any given time, about 20 to 30 percent of the world's total compute is idle and could be allocated for trivial amounts of money to cover the data center cost, and there just are not system structures and incentives to get companies to do that. What we found though is that when you go through the process of talking to the humans who compose those companies, that can change. So we found quite a few companies to have been extraordinarily generous when when these opportunities are presented. Think, you know, in particular have to shout out Dell Computers. I I have to say they've been angels, they've really been saints. I would even say like they have made patients and researchers cry with joy for some of their extreme generosity. And they're not the only ones, but but yeah, it sort of is a thing where starting from that original goal of activating you know a couple hundred engineers, you know we're at this point getting resources to you know like thousands of projects and you know trying to build a community. And our hope is that like we can build a system that will be democratically administered that will allow the world's medical experts to allocate that idle compute. We estimate that about 5% of the world's total compute, let's call it a tithe, could make sure that there's never a rare disease researcher who does not have access to those critical systems. And yeah, that's kind of like the North Star we're following is a moment where, you know, when it comes to getting those resources in the hand of people who need them, it's not insurance companies making the decisions, it's you know, doctors, patients, and patient advocates.

That's awesome, man. And yeah, I mean, that's a crazy stat, just like 5% and like every rare disease researcher can have compute to help them. And you spoke about it or touched on it briefly about the different types of diseases you're researching. I have a question from the crowd I want to bring up on the topic, particularly for Parkinson's and if there's anything that you've seen or you've been you and your team have been researching around it for Tulolu.

Yeah, oh man, I'm I'm so sorry to hear that. I I gotta say, you know, Narb, I am like a big softy. I I think love contributing to medicine, it's I think been something I've wanted to do my whole life, but really like have an admiration for doctors because like it hurts me to even thinking about what a lot of patients go through and, you know, being a patient myself, like it brings up my experiences. And yeah, Parkinson's is such a tough one. A very close person to me has been struggling with that and is a person who actually has a pretty famous career, like working with their hands. So anyway, man, I'm just sending love to you and thank you for the question. Parkinson's is a tricky condition. Parkinson's sort of happens inside the nerve cells. The nerve cells are just a very complicated world, and when you're kind of looking at the dynamics, it's almost like a murder mystery, like there's so many things like floating around. So for example, there's something, Narb, that can form in a cell called a lipid droplet. And a lipid is a type of protein that can serve as like a fuel, it's kind of like a fatty kind of thingy for different cellular processes. So in many cells, it's a natural and normal thing for lipid droplets to form and it's, you know, kind of like putting some stuff in the pantry for the cell. Neural cells like normally don't have lipid droplets. And so like it is an unusual thing we observe in Parkinson's patients, large number of lipid droplets. But yeah, we're kind of like I think there's controversy, like do the lipid droplets cause the disease, are they a byproduct?

And so, you know, that's all to say like right now you know our foundation is working on a compound, it's an autoimmune condition, compound treating an autoimmune condition, and that condition happens because there's like two or three different kind of proteins or compounds that are interacting in a negative way. And yeah, it's pretty likely that Parkinson's involves a large number of biomolecules, like and potentially like extended processes, that are interacting in a fairly complex way. And then again, all of that is is me speaking as like the computer guy. I personally actually hope and am expecting that our next research sprint will be on Parkinson's and a couple related neurodegenerative diseases. In particular, we're going to be looking at what's this is a kind of a crazy thing, Narb, have you ever heard of a human-fly chimera?

No, I can't say that I have. Sounds something out of out of fiction.

No, I know, right? And it's so funny, a fly actually just flew in front of my face just as I was saying that. And listen, everyone in the audience, like next time you're going to swat a fly, don't, like you owe that bro because yeah, like we kind of got very lucky, flies have a nervous system that is remarkably close to ours. And, you know, if we want to test medications, explore treatments, having a model organism where you can test a treatment and then see what happens and, you know, if the life cycle of that organism like the fly is is short, then you can actually go through many generations quickly. So yeah, like flies are very important model organisms, but they do not have human-exact nervous systems. You can give them human-exact nervous systems. So you know, it's sort of funny, Narb, like you're probably used to navigating a big GitHub repo and you see the diagram of like all the different modules. And that's like actually a thing that exists for our genomes too. So when you kind of look at something called the interactome map of a fly cell, you get this thing where there's kind of like this component of the cell is in a box, this component's in a box, each component has different genes that code and produce different proteins. And so it's almost like yeah, it's like the GitHub repo for how the cell works. And what they can do is actually use like a technology like CRISPR to remove the fly genes and put the human genes in the fly. And actually, they can actually give a fly a human bio-exact nervous system for for testing medications. And yeah, we our next project will be looking at using those humanized fly models to do some neurodegenerative disease research.

And then last thing I'll say on that research and before actually giving something that might be helpful like on a more immediate basis for your father, we're at this early moment where AlphaFold 3 and similar models are giving us our first kind of like, you know, think of it as a grainy black and white picture of what would be called the interactome. So not just what a single protein is shaped like, but how does a protein, a molecule, another biomolecule, how do several things actually fit together? And then that technology is is just in its early days. It will continue to scale, gonna need a lot of compute, gonna need a lot of data, but but I sincerely believe like to a sort of would bet on it level that we will soon be able to peek into the the cells of a Parkinson's patient and figure out what what's happening and stop it. So you know, for for you and your father, just please, um, you know, have hope and seriously thank you for, um, you know, being here, being part of the conversation. Last thing I would share though is there is a new treatment modality that a very good friend of mine helped pioneer at UCLA, and it's kind of interesting, it's actually using focused ultrasound on certain brain tissue. So like literally using sonic waves to sort of like you they have like emitters that go on some different parts of your brain so that the sonic waves only like intersect with a certain intensity at like certain tissues. And it has provided really remarkable symptoms relief. Patients who have not been without movement for years like will often have six-month periods with no movement. And I think the kind of kind of name for the treatment would be like neuromodulation, acoustic bioacoustic neuromodulation. You know, again like not a doctor, just just the computer guy, so like you know would say this is more something to maybe you know talk to your father's physician about, but have seen it to be really a modality that's growing and seems to have like enormous impact on Parkinson's patients.

Awesome. Yeah, wonderful answer and hope that that helps a little bit to Tulolu and yeah, wishing your father all the best. And I guess we are running a little short on time and I wanted to give you an opportunity to speak on any other research Rare Compute might be doing that you would like to touch on. I know Rare Compute has recently gotten a big grant from Poseidon DAO to fund some of the fun stuff you guys are researching. Is there anything you want to bring up, touch on? Oh, you're on mute, Stanley.

Oh, thank you, sir. Well, yeah, first of all, like I mean I should have remembered if we thank Dell, we got to thank Poseidon DAO, they're really special group. There's a field and it's I guess the field that I'm in, precision medicine. And precision medicine kind of starts with the idea that we should ask more of of how well we see our our patients. So it's a very data-driven approach. We try to generate as as much data as we can about the patients and then we kind of put the pieces together. And the thing is, it turns out when you do that for one patient, you you do get better outcomes for the patient, but you also increase the health of the total ecosystem of data because we all share so much of our biology that things we learn in one area apply to other areas. And then this is a place where for DeSci it's like there's places where the ecosystem of health can be improved with precision medicine, but it might not make sense for a hospital to pay for it. And so you know many things that are important that would save us money in aggregate across all of society that would be good for our patients, they don't get funding. And then Poseidon DAO is like specifically existing to target that problem, so they're looking to you know allocate resources to projects that are exploring, you know, decentralized and non-traditional ways to bring solutions to patients. And yeah, we we were so excited to be a recipient of their their first grant and we are I think just about to successfully complete the research.

We have our I think it's going to sound weird but our our first hundred thousand molecules were successfully generated. These are molecules that will be attempting to treat a condition called myasthenia gravis. This is an autoimmune condition where your own your antibodies start attacking your neuromuscular junction. And Narb, it's so funny my friend because like often I have to explain a lot of stuff about the neuromuscular junction, but I I know you were a fellow bodybuilder so I'm sure you have that creatine container on your kitchen with the little diagram. Indeed, indeed. That's the neuromuscular junction. So that's like one of the things creatine does is it helps the interface between our our muscles and the supporting neurology to kind of communicate fluidly. And in myasthenia gravis, like you almost could consider it the anti-creatine. One of your antibodies like starts attacking those key proteins and your muscles kind of go haywire in quite a few different ways that are really really terrible. Um, we think we have a a molecule that that should actually grab onto just those antibodies and deactivate we hope enough of them to alleviate symptoms in the patients. And yeah, in in about a month we'll be well first we have to filter those 100,000 down to about 40 or 50. Then we'll test them in the lab, which will produce a lot of data about how they worked, which ones worked. That data will be fed back into the system that generated them and they'll be, you know, regenerated. And then we'll actually go through this process a few different times. And we're hoping that, you know, they're already looking pretty good, but but after a little bit of shaping with this kind of new paradigm which is called wet-dry active learning is what they're calling it. So the wet is the biology lab with the petri dishes and the dry is is me over in the data centers. And yeah, the idea is that a key part of this paradigm is there has to be kind of integrated communication between those those two stakeholders.

So yeah, it's a really cool milestone. Additionally, the traditional paradigm in drug discovery starts with like these these very large databases of compounds that the pharma companies have. And they are very like jealous guardians of these datasets. So there is unfortunately a lot of really important research that like doesn't start because the pharma companies just won't won't open the won't open the pantry, so to speak. And yeah, our method is is very different. You know, we if you want to learn a little bit about what we're doing, if you look at the Nobel Prize from last year, one of the recipients was David Baker at Baker Labs, University of Washington. His contributions to protein informatics like, you know, can't be overstated both in terms of specific research impact, but you know, as I alluded to, getting the stakeholders that need to contribute to these projects like pull in the same direction ain't easy. And Dr. Baker is just, you know, you hear from everyone who interacts with him that he's not just a brilliant scientist, but he's a kind leader, you know? And I just love to hear that. But anyway, he created this thing called RFdiffusion. And if anyone's used like Midjourney or Stable Diffusion or, you know, DALL-E, those are diffusion models. So you know, they learn the connection between some textual data and some images, but it turns out the structured data of the images you can learn whatever structure. So you could actually learn protein structures. You could actually like have a model where it creates the protein you tell it to create. Now the way you tell it to create the protein is a little non-trivial, you're not just kind of prompting it, you're using other biological data. We actually like in and more broadly the field has grown from just using one model to like really using multiple models that are each shaping and configuring a different part of the protein and, you know, there again it's like pretty interesting coordination because like, you know, we often have to start like because we need the biologist to help us configure all the models, explaining to the biologist what a model is, right? Like it's a pretty fun like teamwork kind of kind of thing.

Punchline, though, is yeah, we're we're very close to successful completion of this first molecule with this new technique that doesn't look for the needle in the haystack, it makes the needle. And yeah, we we've done it on I think what would be a bit of a shoestring budget. The Poseidon DAO grant was about 100k. Pretty funny, we actually got roasted pretty hard for asking for too little. And honestly, it was like so like warranted, you know? Like it's kind of like very true that typically a project this ambitious would be looking for like at least $5 million to operate. But we really wanted to prove a point that there are a lot of projects that could happen if we connected the dots. And so yeah, in this case, we were able to find and piece together the things that were needed to follow this new approach. And these were different resources than you needed for the traditional approach. We were able to work with engineers who were just excited to get a chance to contribute to something that had patient impact. And so, although in retrospect I do think we should have asked for, you know, a little more because I'm behind on sleep, I think we're gonna have this successful way to tell a story of, you know, a new approach. And then, you know, my hope is that when we take something like that to, you know, stakeholders, when we have patients helping tell the story of like, you know, why aren't we getting these computers working on these kind of projects? Like, very hopeful that'll be what we need to win people over. But yeah, you know, another thing that helps, Narb, is a chance to tell the story here, you know, to your audience, to you, and also just for myself, you know? It's like a pretty challenging thing have taken on. You know, is it a time where, you know, there are a lot of different things you can get involved in as a machine learning scientist. So I think there's definitely people who think I'm a little stupid for for working on rare disease, but you know, even just a thing like connecting to someone who's a patient and, you know, if nothing else hopefully expressing some empathy and just being part of, you know, your experience going through that as as like the supporter of a patient of a loved one, you know, it sort of like as a again, patient myself, it really is meaningful, Narb. So seriously, bro, thank you so much for having me on, it was a lot of fun.

My pleasure, man. And yeah, those people who are questioning your decision for sticking with rare disease, yeah, they they can go away for all I care because the work you're doing is wonderful, man. You're an amazing person, very brilliant, and I'm really looking forward to seeing how your research plays out here and going forward. Yeah, that's amazing, man. Yeah, yeah, for sure. I feel like we could probably keep going for another couple hours at this rate. But yeah, unfortunately we're at time here. Stanley, thank you so much for taking the time out of your day to come on the show and yeah, definitely you're more than welcome to come back on and give us the full scoop on the research and results.

Yay! Oh, man, it was so good to see you. And you know, Narb, I I will as well. But you know, get a couple sets in on the bench for me, okay? And then next time I'm telling you, bro, we're taking you to Muscle Beach, okay? Hell yeah. Yeah, we'll we'll go scout for Arnold. Hey, we'll see you, sir. 100%. Take care, sir.

All right, hope you enjoyed today's episode of DevNTell. Hope everybody has a very happy Friday, happy weekend wherever you may be. And we'll catch you back here for another great episode next week. All right. Cheers.