The following is a rough transcript which has not been revised by The Jim Rutt Show or Lee Cronin. Please check with us before using any quotations from this transcript. Thank you.
Jim: Today’s guest is Lee Cronin. Lee is the Regis Professor of Chemistry at the University of Glasgow and the CEO and Chief Scientific Officer of Chemify, a recent startup in the area of automated chemistry. Welcome, Lee. Welcome back to the Jim Rutt Show.
Lee: Yeah. It’s nice to be with you, Jim.
Jim: Lee was on relatively recently with Sarah Walker back in episode 100, where we talked about time as an object, and we also got quite a bit into Lee and Sarah’s work in the area of assembly theory. Well worth checking out. Today, though, we’re going to talk about Lee’s business life. What was it about a couple of years ago you started Chemify?
Lee: Yeah. About three and a bit years ago, the first employee started on January 3, 2022.
Jim: Ah, okay. Very interesting. And I have dug into it a fair bit over the last couple days getting ready for the episode. And, man, it’s a quite interesting company. Why don’t you start with essentially your top-of-the-line premise on what you’re doing?
Lee: So the premise is to build chemifarms, like server farms, that can turn code into molecules to make chemicals using a programming language so we can automate things much more quickly. So it’s almost like a production line for molecules that may go into drugs or materials. And it’s not just a production line. It’s a discovery line. So literally, what we’ve done is we’ve taken all of chemistry and built a programming language and a series of systems that will allow us to dial a molecule. And so our customers come to us and say, “Hey, can you make us molecules for this protein target or that are this color or have this physical property?” So Chemify’s aim is to be the AWS for chemistry. You come to us with a chemical problem or specification, we will design molecules to it. That’s the premise.
Jim: Now, essentially you use that analogy because at least today, as far as I’ve been able to ascertain, your model is not to turn the technology loose to the customer, at least not fully. So that’s not really the AWS model. And so far, it looks like you’re mostly collaborative consultative work. Do you intend to actually move in that direction of being a full buy-it-by-the-cycle stack for chemical discovery and synthesis?
Lee: Yeah. No. I would say we are AWS. You think about AWS. AWS is a cloud where you basically deploy your data on that cloud and you buy the services and it works. What’s the analogy with Chemify? Well, we have our cloud rather than being disks in a warehouse is chemical robots in a warehouse. And so then you basically tell us about the molecules you want, and we instantiate those molecules and can make them. And there’s reasons for doing it all within our kind of infrastructure simply because we provide security, safety, and reliability. All the things like, I’m sure AWS wouldn’t let you take part of the servers and bunk it in your house because AWS promises 99.99999999 percent uptime. So from that point of view, I think we’re strictly analogous, and that’s because chemistry is hard, and you don’t want to have people owning their own robots to do it, unless they’re in a research setting or something. So that’s kind of—does that make sense?
Jim: Yeah. I could see that. Yeah. If you focus on the robotic chem farm, then the AWS analogy is pretty good. Though some of the vendors like Microsoft and their Azure cloud, they’ve also offered local versions for corporations that are worried about IP, etcetera, that can interoperate with the cloud, but they want to have their stuff within their firewall. You may find that model is something that particularly the big pharma and chemical companies may be asking you about at some point.
Lee: Well, one day—I mean, right now Chemify is less than four years old. I mean, we have almost 140 people in the company now. Someone pointed out to me the other day, our key product is not molecules, not robots, but factories that make molecules. And, of course, we will be building chemifarms next to our favorite companies that say, “I want my own chemifarm,” and that will be one model. But it obviously will be a different cost. But that’s kind of exciting because in the same way that OpenAI has its own data center or at least has a data center somewhere, I can imagine that Novartis or Pfizer or Eli Lilly would say, have their own private chemifarm for mining chemical space and generating assets. But, actually, the security is fine because we encrypt everything anyway. The Microsoft Azure model, I’m not so familiar with, about having stuff because, obviously, Microsoft and Amazon are slightly different companies. But for us, right now, what we need to make sure is that chemistry shifts from manual to automatic, and this is a radical change. If I was to say, you know, one sentence, what does Chemify do? We do chemistry automatically, and that allows us to scale. And if we can’t do it automatically, then we can’t scale. So every time an artisanal chemical chef tries to say, “Oh, hang on a minute. I want to do this by hand.” We say, “No. No. No. Use the programming language. Use the system.”
Jim: Yeah. Indeed. And as I was digging into it, I found some of the things you’ve done, which is an attempt to build a full stack as it appears. Looks like you started or early in the process at least, you created a programming language. How do you pronounce that? Is that
Jim: Yeah, indeed. And as I was digging into it, I found some of the things you’ve done, which is an attempt to build a full stack, as it appears. Looks like you started, or early in the process at least, you created a programming language. How do you pronounce that?
Lee: ChemChi—the Greek letter kai, like an “oi”—XDL, and it started off as a description language. The kai is for chemistry, and it turned over the years from a description language to a programming language. For those people listening who want to know, the difference is super important, that transition, because if you can make a formal programming language that is formally verifiable, then suddenly you have a way of making sure that your code is portable. With that portability and executability, you have interoperability and all the beautiful stuff you get from writing Turing complete programs. The brutal reality is that chemistry was not Turing complete until about a couple of years ago when we made it Turing complete in my laboratory and then in the company.
Jim: For the benefit of the audience, could you explain what you mean by Turing complete in the domain of chemistry?
Lee: This is a super interesting question. Let’s imagine what a Turing machine is. A Turing machine is a very simple abstraction of computation that uses a one-dimensional tape, a head, and a lookup table. By a simple set of rules, you should be able to simulate any program or any computer running a program on that Turing thing. You could say that you can’t get any more fundamental than a Turing machine. That means you could run Microsoft Windows, BBC Basic, Linux, or whatever on these and all the programs.
What is a chemical Turing machine? Well, a chemical Turing machine is an implementation of the Turing machine that allows you to run any chemical process to make any molecule. I had the intuition this was possible for many years, but I only really finished the mathematical proof three weeks ago. On the tape, you have cells rather than writing a one or a zero or space—the cells can be full of stuff for the reaction, active like for the chemical reaction, or empty or blank. With those three types of cell in the tape and minimal rules, you can actually take any chemical program, any synthesis of any molecule, any drug, and turn it into that via four primitives.
In computer science, the primitives are relatively simple in the machine code. In chemistry, they’re even simpler. There’s only four: add matter, subtract matter, add energy, and remove energy. When I realized those four parameters were the ones that you needed to use again and again, it allowed me to complete the abstraction. That’s a very long-winded way of saying that we now have a universal way to represent the synthesis of any molecule, and that can run in a chemical Turing machine that then can be put onto qualified hardware. That suddenly opens up the standard so that people can interoperate and collaborate.
Jim: Now should I take that to mean that you have discovered that chemistry can be defined strictly in the classical realm and you don’t have to ever consider the quantum realm?
Lee: No. You can do quantum computation as I call it. That’s really interesting in itself. With standard computation, classical, we use the standard classical ways of manipulating chemistry and accept, you know, it’s like you take a proportion material, heat it up, cool it down. If you can do quantum, you could suddenly say, “Oh, hang on. I’ve got this chemical and this chemical, and I’m going to cause a chemical reaction to happen by putting in a quantum of energy.” Or shaping some energy so I can efficiently cause the reaction to occur. But actually, kind of really interestingly, there is no real difference between quantum computation and computation. They are the same phenomena.
Jim: Well, that’s actually an important result in and of itself because we think one of the things I often think about is the tension or tension potential gap—still not perfectly quantified—between the Church-Turing thesis and the Church-Turing-Deutsch thesis. That there is a need for certain physical phenomena to consider the quantum realm. And chemistry is, of course, right close to the gap. When you get down into particle physics, you’re certainly in a quantum realm. Chemistry is on the edge. So this is very interesting to me kind of theoretically.
Jim: Well, that’s actually an important result in and of itself because we think one of the things I often think about is the tension or potential gap—still not perfectly quantified—between the Church-Turing thesis and the Church-Turing-Deutsch thesis. Right? That there is a need for certain physical phenomena to consider the quantum realm. And chemistry is, of course, right close to the gap. When you get down into particle physics, you’re certainly in a quantum realm. Chemistry is on the edge. So this is very interesting to me theoretically.
Lee: You can make computation quantum, but in the same way, quantum computing has kind of become nonphysical. The reason why I wrote up the math is that David Deutsch asked me to. David Deutsch has a theory called constructor theory and part of the constructor theory—we won’t go into it here—but he likes catalysts to affect transformations. And he saw that computation and computers provided a different framework to cause transformations. So he encouraged me to explore the Church-Turing computation thesis, as it were. And of course, you can probably take the Church-Turing-Deutsch computation thesis together.
But with quantum computing, we have this ability to use many qubits for encoding a problem and potentially reading it out if you can overcome the eta error catastrophe. There’s not the same analogy in chemistry because here we’re talking about using quantum focal coherent chemical control. What that means is if you do your quantum chemistry correctly, you should get very high selectivity and no waste products. So the difference between quantum computing and computing could be like no waste, zero waste, absolute transformation. That’s a slightly different thing, but I’m still working on that. I can barely scratch the mathematics for chemical computation.
Jim: You better get Deutsch to help you with the quantum if you need to get there. Interestingly, I do talk to people in the quantum computing industry fairly often, and quantum chemistry is an area they’re very interested in. One example reaction comes up again and again, and that’s synthesis of ammonia from nitrogen in the air. Right? It’s a gigantic industry, I think 100 billion dollars a year or more. And there’s reason to believe that there are more efficient reactions and catalysts that have not been used. It’s a classic example question on the utility of quantum computing, if any—can it be actually used to improve significantly the synthesis of ammonia? Any thoughts on that?
Lee: That’s not quantum computing. And this is where the slight woo-woo in quantum computing I find really quite perplexing. The big problem with quantum computing is we don’t know about error correction and encoding information into qubits, and allowing those qubits to basically give you information back that you then decode and get an answer. That’s what I would say is quantum computation.
But the people that are selling this stuff are going, “Oh, you know, maybe that doesn’t work right now. There’s a gap in the software. We don’t know what it means, but we can quantum compute chemical reactions.” That’s not what they’re doing. What they’re actually doing is taking quantum simulators to solve the problem of nitrogen and hydrogen to ammonia. You do quantum theory, you look at what’s called the transition state to put the hydrogen on and see how you can lower the energy. And that is a quantum simulation. That can be done classically.
The way you do it classically is you instantiate it in silicon and you flip a load of matrices. It’s quite intensive and it scales badly with the number of atoms. But what if you could just take some quantum elements that behave like the quantum system of ammonia and tickle them so they vibrate and you can do a simulation? I would call that some kind of quantum simulation, but I can do that all the time without a quantum computer. I don’t need a quantum computer. So when people were selling me quantum computers for chemistry, I’m like, I can just do a reaction under quantum control, and I don’t need to use a 10-million-dollar instrument at liquid helium temperatures. Thanks.
As a chemist, I maybe know too much chemistry and not enough computer science. I guess that’s a long-winded way of saying quantum computers can also be used for simulated annealing and quantum systems. That’s super interesting. But it’s not a computation in the conventional sense. It’s more of an emulation.
Jim: Of course, that was Feynman’s original motivation to think about quantum computing, was to use it for that. So anyway, we’ll move on from that. That’s a very good answer. I like that. It’s going to help me a lot in my conversations with people who, I will say, sometimes seem to be overhyping their quantum computers, shall we say. With respect to the programming language, I found it online, and I will have a link to it on the episode page at jimruttshow.com. Also, for those knowledgeable enough to play with it, Chemify has a Chem IDE that lets you actually experiment with the programming language in a web-based IDE. Does that actually work these days?
Jim: Of course, that was Feynman’s original motivation to think about quantum computing, was to use it for that. So anyway, we’ll move on from that. That’s a very good answer. I like that. It’s going to help me a lot in my conversations with people who, I will say, sometimes seem to be overhyping their quantum computers, shall we say. With respect to the programming language, I found it online, and I will have a link to it on the episode page at jimruttshow.com. Also, for those knowledgeable enough to play with it, Chemify has a Chem IDE that lets you actually experiment with the programming language in a web-based IDE. Does that actually work these days?
Lee: Yeah. So that was pre-Chemify company, Chemify kind of organization. One of the things I set up deliberately to do, because I want to change the world of chemistry—there’s two ways to make the world change the world of chemistry. First one is to change the culture, and the other one is to change the tools. And to change the tools, change culture, train new people, make new molecules, do the capitalist thing, everyone is happy.
But I realized making a company wasn’t just enough. We had to start an open tool movement and get awareness in academia. So that Chemify IDE was right from the Cronin lab. Can you program chemistry to get people used to it? And you can use it online. It gets you used to it. What Chemify uses is a highly advanced version of that, which is optimized for the hardware that we’ve developed in-house. But the nice thing, because it’s Turing complete, it’s compatible with academic things.
One of the things I’ve always got is this conflict between trying to make sure we get a proprietary Chemify that can work fast and making sure that academia and society as a whole can benefit from this. And actually, you can do both at the same time. We’ve shown that again and again. And so I made the decision to set both balls running at the same time because I think that the movement in chemistry, the culture change and the ability to make chemistry automatic is something that’s going to cost billions of dollars and take tens of years. The quicker we can get people changing their mindset and building companies and discovering drugs and working across academia, the better. And if we have robotic systems in universities, we might be able to have centralized laboratories where people around the world who don’t have access to labs can do a chemistry degree online, and they can just use this Turing complete flight simulator.
Jim: That’s interesting.
Lee: Yeah. Exactly. It’s like, I think it opens up chemistry. So that’s kind of my dream, if you like, is to make Chemify a 100 billion dollar company, make sure the programming language permeates all of chemistry like chemical drawings so people are able to swap designs more quickly, and that we change the world because right now, there’s more molecules possible than there are atoms in the universe.
Jim: Indeed. Though you’re dealing with Planck’s principle, right, which is science progresses one funeral at a time.
Lee: Well, also there’s Cronin’s principle which is annoy everybody so much that they will be cognitively disimpaired for a bit and then you can convince them. And it starts with youngsters, of course, and also old young people. And what I mean by old young people are just people that are willing to change their mind and try something new. So it doesn’t have to wait on everyone’s death. It just needs to be useful. And that’s what I’m excited about. Because at the end of the day, by me being forthright with people, it allows people to tell me why it’s wrong.
For many years, people told me I was just making chemputation up. There wasn’t a thing. And I’ve really struggled to explain to them that you want a general purpose programmable computer, not a test tube. And I think now with the formalization of the language and the standard, it would be kind of limited. So I’m really excited. I mean, also, it’s really hard. I had to learn some computer science. I had to build this up, and David Deutsch was like, “It’s obvious. Just do it.” I was like, “Thanks, David.”
Jim: Interesting. Yeah. And as always, you know, show them is the best way to convince people.
Lee: I was very fortunate in Glasgow. One of my colleagues is a formal methods computer scientist, and she said, “You know what? There’s something here” because her research helps you formally verify control systems. Despite all the bad press that certain airlines get, I’m very happy to get on an airplane made by Boeing or Airbus knowing that their software is formally verified. It’s not written using ChatGPT. That means that you look at the decision trees and you’ve got redundancy built in and it’s a very, very well characterized system.
Imagine a world where everyone has computers, but if the code is wrong, the kitchens or garages or laboratories catch fire. This is the danger that we have if we don’t have a programming language with a standard ontology. Right now, people are thinking, “Oh, we can program chemistry using ChatGPT.” Well, the ChatGPTs literally make stuff up. And if you make stuff up, there’s three possible answers to chemical reaction: you get a fire, it doesn’t work, or you get the molecule. You want the latter.
Jim: Good thought. I love this idea of chemistry now being something that could be done virtually. One could imagine, if one was a theoretical chemist and the full Chemify stack existed and was cost effective, it’d be interesting to hear your ideas about how you might provide access to the university world, the academic world. One could get a PhD in theoretical chemistry or even practical chemistry remotely if they had access to all the tools.
Lee: That will happen 100 percent. And it’s a question of who makes it happen and how quickly. I mean, Chemify was named after Spotify. And why did I name it Chemify after Spotify rather than Chemical Napster? Well, Chemical Napster—or Napster was just when people ripped all the music and put it online. And obviously, there was a big copyright class action lawsuit and Napster died. But the people that won the music streaming wars learned to not break copyright, to pay the creator.
Wouldn’t it be cool if a new professor in Stanford or in Paris or in Tokyo or in Timbuktu came up with a new code to make a drug really efficiently? And rather than publishing it in a journal where no one reproduced it, they just put the code on a GitLab or Chemify’s lab and you can download it. And the more downloads you get, the more plays, the more kudos you get—or maybe even money because it takes a graduate student many years to reproduce some of the literature sometimes. The idea behind Chemical Spotify was to allow people to copy each other more quickly, and you get more collaboration. The developments we take for granted in computer science and engineering are facilitated so much by these collaborative tools that simply don’t exist in chemistry. And KIDL computation in the computer is the first attempt to do that rigorously.
Jim: I can see how that could very much accelerate the social sharing of knowledge and the reuse of knowledge. Let’s move on to the next part of your stack, at least as I understand it. So we start with the language, we then have some programming tools, and then the other thing that you guys have created is your Chemify database and reaction library. Why don’t you tell us about that?
Lee: Yes. So this is kind of part of the vision. Before I built the company, this actually happened in the pandemic. Accelerated a lot. Pandemic comes, and I’m building these robots before the pandemic, and everyone’s like, “Lee’s building robots. That’s really annoying.” Pandemic comes, and suddenly, I have to send fifteen organic chemists home. And I was trying to convince these organic chemists to learn the programming language, and they were quite resistant because they loved doing organic chemistry. They’re like, “Lee, you’re taking away the joy of organic chemistry and making me program, and I don’t like programming.” And I said, fine.
Suddenly, we’re locked down, and then I convinced the university to allow me to unlock my lab and open it up again because I have robots. I said, look—we just have two people come in a day. They load up the robots with chemicals, and they go home. And then people log on to the robots and do the chemistry. That was the idea. So what the people had to do was actually develop the programming language by trial and error, reading the literature, and checking and coming up with the database.
The idea would be we take some molecules that are well known but hard to make because you want to convince the naysayers, put them in the robots via the program, run the program, run the robot, and check that the molecule you got out is what you expect. So you have this KIDL entry you put in the database. You run it, and then at the end, you get an outcome and you put that outcome in the database and there’s a version. Now let’s say you only get a 20 percent yield and it’s pretty poor quality, like impure. That’s good. You’ve got that labeled in there, but then say someone goes, “Ah, Lee doesn’t know what he’s doing. Use the wrong catalyst or too high temperature.” Change that, and then you run it again, another version. And this time, you get a better yield and it’s higher quality. So what you can have is a kind of version control for the chemical reactions all open for the people to see, and that’s where that Chemify database came from that we were running connecting to the robots and so on.
Jim: Very interesting. How does that relate to things like the open reaction database that people in the chemistry world have been trying to get rolling for some time?
Lee: So I know the ORD very well actually, and there is a secret website that I have. Let’s see if it still works called Syntex—syntex.com—and it actually interfaces ORD to KIDL. So those geeks who are listening, go there and have a play. Now ORD is a manually curated database of manual chemical reactions, and they should be slightly higher quality than the average. And so what we did is used a lot of the ORD data to build code that we put in the robots. And one day, I would like ORD to also take the role or be a part publisher with the academic side of the KIDLs. The academic Chemify. Whether they’ll do that, I don’t know. The only thing I know how to do is to build stuff. If I build—I’m not the most diplomatic in the world.
Jim: Oh, you’re not. Who would have guessed that?
Lee: I do my diplomacy by actually doing what I said I’m gonna do. So shock and awe.
Jim: Yeah. You’re the Bismarckian style of diplomacy, not the Metternich style.
Jim: Oh, you’re not. Who would have guessed that? Right?
Lee: I do my diplomacy by actually doing what I said I’m gonna do. So shock and awe.
Jim: Yeah. You’re the Bismarckian style of diplomacy, not the Metternich style. Right?
Lee: Exactly. And so I don’t know where that will land. I think the R&D is integral, but the thing is we are not generating data in chemistry fast enough to make an impact. And there is a temptation with AI and synthetic data for people—very entrepreneurial computer scientists with a bit of chemistry background—to generate synthetic data, and that is not gonna help us. Because the synthetic data, the simulated data is not real data.
Jim: That’s where I wanted to go next, actually. Do you have a reasonably effective means to essentially reverse engineer ORD data back to your ChIDL form? Can you take an entry—
Lee: Yeah. No. It’s not a reverse engineer. We have a schema which will allow you to take any published chemical recipe, and that’s what the ChemIDE does. You can type in in words, you know, do the blah blah blah and press convert, and it will convert it to an academically compatible ChIDL. Interesting. And that’s really good because it formalizes the prose and turns it into a structured language. So ChemIDE is almost like a linter, a Python linter, but a chemistry linter. So what you do in ORD is you take your ORD and you copy and paste it into the IDE and press translate, and it will produce a ChIDL code. And where there are missing pieces of data, like maybe some temperatures or solvents or something, it will make suggestions from its knowledge base.
Jim: That’s extremely interesting. Have you done that at scale? Have you taken large amounts of the ORD and turned them back into your code?
Lee: Yeah, all of it. There’s a few hundred thousand, let’s say, that we’ve done. I’m kind of working on a new academic version of it. The problem is my day job is CEO. So right now, I think it’s very important that I get Chemify to a critical mass. And I don’t have anything against academia, but the problem is the opportunity is that there is such a critical problem getting access to molecules outside of academia. It’s all done by manual labor that I really need to solve this problem. And the only way to do is a hyper high-growth startup that is singularly focused on turning code into molecules efficiently at low CapEx. This is really the most important thing. People spending vast amounts of money on chemical robots, and they can’t do any chemistry. And I realized I had to find the sweet spot—basically be faster, be cheaper, and be scalable. And they’re the three and I was like, holy cow. These are the three elements of capitalism. I better be a capitalist and let this run. And I think that the company will expand explosively. We’ll build Chemifarms all around the world, and you probably know this. Our first Chemifarm opened three weeks ago in Glasgow.
Jim: Indeed. Well, that’ll be our next topic is talking about the robotics aspect. But before we go there, you mentioned in passing synthetic data. And one of the things that that raises is rather than always having to use even the Chemifarm to produce the molecule, where does the world of chemical simulators sit these days with respect to being able to have a faster loop of going from code to simulation? I asked that as I was involved with two startups, both of which were successful, that use simulators in the loops, machine learning, and genetic programming to optimize analog circuit design. The whole reason that the business worked is there’s a very high fidelity, reasonably efficient that we had to build our own vast data centers of analog circuit simulators. Are there simulators yet in chemistry and organic chemistry in particular good enough that there are domains where simulator and the loop may actually be a useful step in the process?
Lee: So the quick answer is no. They should be. And that’s because chemistry is very analog and the data is pretty bad. But that’s one of the things that Chemify is going to do. There is a finite number of reactions and finite number of building blocks. And once we’ve done those—or a finite number of reaction classes rather and building blocks—and once we’ve done those and we go to a certain depth of chemical space, we’ll be able to simulate and validate the simulation of that chemical space. So let’s say do the first 10 to the 20 molecules. That event is not very far away from Chemify.
Arguably, Chemify’s current simulatable chemical space—and I’ll come back to what we mean by computational chemical space—means the molecules in that space can be encoded, and you have a very high chance, if not certainty, of being able to make them in the lab for real. I’ll say that last sentence again: to make them in the lab for real. The problem with all other chemical simulators is you cannot make the molecule that you simulate in principle because you don’t know the route.
What Chemify does is we have some secret way—it’s not that secret, you can work it out—of making sure the molecules we invent, we can make. That capability doesn’t exist yet. And once we have enough of that, exactly what you’ve got in semiconductors right now, the ability to simulate circuits, to basically design things. I’ll be able to design a molecule that I think will bind to a protein and be a drug, and I can do it a zillion times. I’ll then take these as assets and encrypt them, maybe make a chem coin like a Bitcoin and put it in a ledger somewhere. And then when someone gets around to making it, we can test it. And then I own part of the asset and so does the person who’s tested it. It’s a way of making an NFT for molecules.
Jim: Don’t do that. Something else in that line of essentially sharing IP is a good idea. Don’t do it as an NFT.
Lee: Yeah, what I’m saying here is you want to give creators the ability to fingerprint their work and then enable others to build on it.
Jim: That in that ecosystem. So I absolutely will do that, but not today. Okay. Good. So I was right that simulation in the loop is an important part of the strategy, but you’re not quite there yet. Correct. Okay. That makes a lot of sense to me. That will make your business even more valuable. And fortunately or unfortunately for the world maybe, simulators in the loop can be kept as trade secrets if not patentable. So there’s some business opportunities there for sure. Though on the other hand, if you want to do good for the world, putting the simulator out into the world would do a hell of a lot of good.
Lee: You know, I think having the simulator out there is the way forward. Academia—because there are more possible molecules than there are atoms in the universe—we aren’t running out of molecules. And I think that having it out there will allow us to create so much value and, more importantly, change the way that we enhance the human condition, cure diseases we never thought we could ever cure.
There is an argument to get this right early rather than later, because the Internet is decaying right now. We tried to do it right by making sure that all the Internet traffic was free and all this stuff, but it’s decaying, decaying, decaying. I think that it’s important that we get these simulators, or at least some prerequisite for them, out there so that academia and open source can use them for good and for evaluating molecules that would never have a chance to be evaluated. Let’s be clear: chemical space, combinatorial chemical space is bigger than anyone can possibly imagine and then some. So it makes no sense to keep that technology secret.
Jim: You’re currently working in the kinds of chemistry that’s interesting to pharma and material science, which are medium-sized chains. What about really exotic stuff like nanotechnology, where we actually have molecular machines? Do you see that as being within the scope of what Chemify will be able to address at some point in the reasonable future?
Lee: Actually, some nanomachines are within reach right now. Just a few months ago, my research group published a paper—I can send you it for the show notes—where we used a computer and a KDL to make a molecular machine. So a machine that made a machine. There are obviously limitations on how big you can make the molecules in the short term, but in principle, because computation automates everything and you have error correction, you should be able to make quite large molecules. I was joking with one person saying, “Hell, maybe I can even synthesize you memories.” Memories are molecules at the end of the day. What new memories would you like?
Jim: Interesting. This touches obviously on assembly theory. Right? So in some sense, good old Lee Cronin and his gang are accelerating the advancement of the complexity of the universe at a far faster rate than Mother Nature herself was doing it.
Lee: The computation paper that’s on the preprint server just now actually uses assembly theory to redefine what a molecule is. A molecule in a chemical sense is a thing—it is a molecule if you can put the atoms together and they won’t fall apart. But that isn’t enough. You need to be able to, in principle, make enough of the molecule to detect. And what assembly theory says is it puts the limits on that. And so I was able to merge them together, further confirming I’ve only ever had one idea, but I just keep exploring it.
Jim: Interesting. Like, see where this touches obviously on assembly theory. Right? So in some sense, good old Lee Cronin and his gang are accelerating the advancement of the complexity of the universe at a far faster rate than Ma Nature herself was doing it.
Lee: I mean, the computation paper that’s on the preprint server just now actually uses assembly theory to redefine what a molecule is. A molecule in a chemical sense is a thing—it’s a molecule if you can put the atoms together and they won’t fall apart. But that isn’t enough. You need to be able to, in principle, make enough of the molecule to detect. And what assembly theory says is it puts the limits on that. And so I was able to merge them together, further confirming I’ve only ever had one idea, but I just keep exploring it in different ways.
Jim: It’s actually a very interesting idea. It’s engaged my interest quite a bit. My most recent conversation with Sarah Walker ended up diving back into it as well. The next part are your robots, right? Your computers. Tell us about that. And how does a poor innocent bench chemist get himself entangled with fucking robotics?
Lee: One of my key research questions was to understand how chemistry and biology kind of interact. And I, in a crude way, wondered how does chemistry become biology? And I crudely thought of it, and this is the wrong way to look at it—it’s the Boltzmann brain problem. But I just said, okay, if the chemistry in biology is a subset of wider chemistry, if I make a search engine for chemistry, I can discover biology.
Jim: Maybe. Maybe.
Lee: Well, that’s what I thought.
Jim: Anyway, continue. Continue your story.
Lee: One is a subset of the other. Well, I then thought, I’ll just build a search engine. There must be software out there, and there must be robots. But to my horror, there was no programming language for chemistry and there was no robotics. So I initially built the programming language for chemistry in 2012, 2013. I went to my research group and—this is something I wrote myself—and said, okay. It wasn’t that complicated. It was literally a glorified system: can you please do what you’ve written in your lab books and formalize it? Like, if you’re going to add A to B, add A to B and not B to A. If you’re going to use 30 milliliters as solvent, don’t use 300 milliliters. Don’t forget your stuff. Just write it down clearly so it’s machine readable.
But when I took that to the group, they were so annoyed with it. They were like, “Why would we program chemistry? We’re well trained. We can read our notes and do it.” I said, “Well, what about if we built robots and we had to instruct the robots?” And they went, “Oh, that’s cool.”
So I started to build these robots, and I used 3D printer architectures because they were cheap microcontrollers. It was possible to start building architectures and graphs on them. That’s how I started. I’ve always been a geek, and the first computer hardware I built in my garage, and kind of encoded and built the initial firmware. Then with my team, we put a very basic computer together and showed that we could add two solvents together and then add two reagents together and then heat it up and do all this stuff. It was really exciting because suddenly in the lab, I could bring in not just chemists now. There were computer scientists, engineers, mathematicians, operating system developers, people who are experts in networks. So the group really grew in diversity and interdisciplinary work, which I thought was really awesome.
Jim: Maybe you could just give a very high level for the layman and for me as well in this case. What is it that these robots do? Run through a scenario.
Lee: All the robots do is they get the chemicals to the right place at the right time, and then they add the right amount of energy and remove the energy, and then they allow the material to be prepared and separated so it can be purified. If you think of chemistry as a series of set steps where you have to make this happen, where you’re literally adding reagents together at the right rate, heating it up, cooling it down, doing the separation, that’s what the robots do. These robots are a combination of liquid and solid handlers, and they have control systems on them for doing stuff. What we’ve done is we’ve made sure the robots are compatible with all the different requirements that chemistry needs because chemistry—the molecules can be dangerous. They can catch fire, they can explode. So you want to make sure that everything is well controlled. We’ve done a lot of work in making sure that works out, and that’s what they do.
Jim: I’d assume they’re not little mobile things running over to the cabinet, opening the cabinet, bringing a reactive reagent out, pouring it into a beaker, putting it on a Bunsen burner. Presumably, there’s a whole bunch of piping and automated distribution of chemicals.
Lee: Yeah. I mean, you could do that, but it’d be very expensive.
Jim: Be kinda silly probably.
Lee: Lots of startups have tried doing that. The real question is, the robotics is almost a red herring. The robotics are guardrails for making sure your chemical code is reproducible. So put another way, what is the minimum amount of robotics you need to make the chemistry reproducible? And that may or may not be something that my top secret company is trying to do.
Jim: I’d assume they’re not little mobile things running over to the cabinet, opening the cabinet, bringing a reactive reagent out, pouring it into a beaker, putting it on a Bunsen burner. Presumably, there’s a whole bunch of piping and automated distribution of chemicals, et cetera.
Lee: Yeah. I mean, you could do that, but it’d be very expensive, right?
Jim: Be kind of silly probably, yeah.
Lee: Lots of startups have tried doing that. The real question is, actually, the robotics is almost a red herring. The robotics are guardrails for making sure your chemical code is reproducible. So put another way, what is the minimum amount of robotics you need to make the chemistry reproducible? And that may or may not be something that my top-secret company is trying to do.
Jim: I got you. I mean, it would make a lot of sense because the vision here is a full stack, right? So one thing constrains the other. The fact that you have this chemical language essentially is a pruning rule for what happens at the next stage. Presumably, the robotics then are designed to operate within the metaphor of the programming language.
Lee: Yeah. And I think that’s the trick here is to basically restrict operationally what the chemists and the robotics can do, but open up the maximum combinatorial space of molecules that you can therefore access by that reliability. And that is a nontrivial process, and that’s one of the reasons why I got so much pushback because all the chemists said to me, “You’re just making stuff up. It isn’t gonna work. My compounds are really complicated.” And I was like, well, you know, you are using the laws of physics and laws of chemistry. I’m sure that we can make a compromise. It’s a bit like saying we can’t build semiconductor industry based on binary. We should use all the different combinations of binary, trinary, whatever. No. Let’s converge on a standard that can do everything. Let’s build a paradigm. Let’s build a technology, and let’s make it reliable.
Jim: Ah, now let me ask you a geeky question, very analogous to what happens in computer science. Now as you’ve actually gone down the stack to synthesis, have you learned things about your language that required you to revise it, that it did not handle certain cases, and so therefore you had to expand the scope of the language? I’d love to hear any stories about that.
Lee: Yeah. Actually, we did it the other way. I basically said to everyone, because I realized I was an abstraction engineer, and a lot of my team got frustrated because I was always arguing with them saying, “No. That’s not a primitive. This is a primitive.” And so to get around that and to get everyone to inspire everyone to take part, we just took the entire Internet of chemistry, if you like, and just coded it all. And then we did it in such a way on a knowledge graph where we looked for common parts and we could remove them out. Because I had assembly theory in my head at the same time, I was able to look for redundancies and reduce and reduce and reduce. And in the first iteration, we got it to forty-two primitives. And I was like, forty-two? Well, that’s great if you’re The Hitchhiker’s Guide, but forty-two still seems a lot.
And I realized that chemists have lots of different words for if you want to remove something—you could be drying it, you could be extracting it, you could be doing all this. And actually, it comes down to subtraction of matter. So if you’re removing water or drying an organic solvent, you are actually subtracting water from the organic phase. So subtract matter. And there’s a whole multiplicity of terms which means subtracting matter. And so it literally has taken me ten years rather embarrassingly to get from the entire literature then to forty-two primitives. And then about six, eight weeks ago, I realized they didn’t have Thor permit. Thor permit is sub add matter, subtract matter, add energy, remove energy. And every other action can be put together as a composite of those four things, and that was all we needed. And I was like, holy cow. This really is the transistor for chemistry. It’s not quite on and off, but it is. Stuff and no stuff, heat and no heat.
Jim: Yeah, at the grossest level, but in actual chemistry as it’s practiced, especially in industry, the most important aspect, at least as I understand it as a layman, is catalysis. How do you capture catalysis in your language?
Lee: Oh, it’s just added as one of the process conditions in the KIDL file. So it’ll be literally add catalyst. And because the definition of a catalyst is a thing that accelerates the chemical reaction and is unchanged in the process, you can add the catalyst, chemical reaction occurs, and at the end, you can remove the catalyst.
Jim: Okay. How does that help you discover appropriate catalysts? Because, again, in industry, for sure, that is one of the key questions is trying to continually refine your catalyst, either individual catalyst or your ensemble of catalysts.
Jim: I mean, yeah, at the grossest level, but in actual chemistry as it’s practiced, especially in industry, the most important aspect, at least as I understand it as a layman, is catalysis. How do you capture catalysis in your language?
Lee: Oh, it’s just added as one of the process conditions in the KIDL file. So it’ll be literally “add catalyst.” Because the pat definition of a catalyst is a thing that accelerates the chemical reaction and is unchanged in the process, you can add the catalyst, chemical reaction occurs, and at the end, you can remove the catalyst.
Jim: How does that help you discover appropriate catalysts? Because, again, in industry, for sure, that is one of the key questions is trying to continually refine your catalyst, either individual catalyst or your ensemble of catalysts.
Lee: Well, we can now make a loop. Let’s do this properly. Let’s pretend we have our catalyst discovery machine computer, and we’ve got two types of matter we’re going to add. We’ve got matter, which is active matter—we want to add them together and want our reaction to go. And we’ve got another type of matter, which is catalyst matter. So what we do is we have a whole load of pots with different matters which we think might be catalysts, and we have a whole bunch of pots which are active matter. And all we do is we chuck the active matter in the active cell of the chemical Turing machine, and we measure: Has anything happened? Yes or no? And then if nothing has happened, what we can do is add another piece of matter, which we think is a catalyst and check and say, has it done anything? Yes or no? And we can use Bayesian processing to basically do that.
I wrote a paper on this a few years ago where I wanted to call the paper “Every Chemical Dog Has His Day or Her Day,” but the team were like, no. What I mean by that is, what is a catalyst? Everything in chemistry can be a catalyst for something. The question is finding out what it is. It’s a combinatorial problem. That’s a way to make a catalyst search engine and I made a few of those. But of course, you put knowledge in. Most chemists don’t start from scratch. They basically have an indication of what type of materials are good for catalysis, and so they put them into the system and just improve it using design of experiments, and that’s indeed what we have done.
Jim: From a business perspective, catalyst discovery and ensembles of catalysts and changing the catalysts have huge economic value if you can change the reaction efficiency even a little bit. Is that today a big source of the commercial interest in your work?
Lee: Yes, I mean, there’s a lot of interest, but the problem is there’s a lot of AI vaporware out there, and Chemify has purposely tried to stay away from it to say the proof is in the doing. And there’s a lot of people out there not only pretending to be able to make catalysts using AI, but also pretending to use the catalysts to do reactions. And that’s not saying that it’s all made up. What I’m trying to say is, look, you need the robots to work. You need the transformations to work, you need the search engine to be a physical search engine. And once you have that infrastructure, you can go to the races. And Chemify is doing that with a few partners just now. And like you say, if you can just improve a catalyst by a thousandth of a percent, if it’s a multi-ton business and it’s high value, you’re talking about millions of dollars saved and generated and energy saved and pollution reduced. So, yeah, I think it’s going to be a huge aspect for us later. Chemify isn’t just doing drug discovery. We’re making molecules so people might do drug discovery. We’re going to look at fragrances, paints, catalysts, lipids, monomers, for polymers. If it has electrons on it, we’re game.
Jim: How does the work with people like AlphaFold impact the domain that you’re operating in? Because catalysis, particularly called enzymes and organic chemistry, are significantly operationalized by their physical structures. And physical structures have been historically very hard to find for long chain organic molecules. AlphaFold now seems to be able to have substantially moved the goalposts on getting to physical structures from inorganic chemistry. Are you leveraging that or is that a different domain?
Jim: How does the work with people like AlphaFold impact the domain that you’re operating in? Because catalysis, particularly called enzymes and organic chemistry, are significantly operationalized by their physical structures. And physical structures have been historically very hard to find for long chain organic molecules. AlphaFold now seems to be able to have substantially moved the goalposts on getting to physical structures from inorganic chemistry. Are you leveraging that or is that a different domain?
Lee: So, AlphaFold—we should take a step back. What does AlphaFold do? AlphaFold has taken the corpus of protein sequences of which they have got structures and used machine learning to turn sequence into structure. That works very well when your query is within dataset. It doesn’t work very well when your query is outside dataset. So AlphaFold hasn’t actually solved the protein folding problem. What it has done is made an incredible tool to generate new proteins or sequences as long as you understand the domain in which they’re working.
Now for catalysis, proteins are great, but they decompose—they’re very fragile. In terms of mass to active ratio, it’s quite low because they’re quite heavy. But when it comes to generating ligands that might bind to the proteins’ potential drug discovery targets, AlphaFold will make a significant difference. It won’t be the be-all and end-all because it doesn’t code for confirmation of physics. AlphaFold doesn’t understand physics. I don’t care what people tell us. There are rules you can encode in.
So that’s a very long way of saying no, AlphaFold hasn’t impacted us very much at all, but it is a fantastic machine learning achievement. In fact, one that was fully understandable ahead of time. I’m not demurring—it’s fantastic. What did AlphaFold show? If you have a very good set of data, you can machine learn on it and use it for a great tool. And I think we can use that concept in catalysis.
I’m not sure whether we’ll be able to make protein catalysts for small molecules simply because—and you may not know this, Jim—but I did my PhD in bioinorganic chemistry where I made protein active site models as catalysts. And, of course, when you make an active site catalyst, it’s smaller than the protein, and it doesn’t have the secondary and tertiary features of the protein. The protein is breathing. It’s doing other things that evolution has selected for. So cutting corners or removing a lot of the mass out there is nontrivial. It will be possible, I’m sure, to solve that problem, but AlphaFold is not the tool for that problem.
Jim: Let’s now move on to some other issues. What is your business model today, and how do you see your business model unfolding in the future?
Lee: You give me money, I make you molecules. You give me even more money, I design you molecules. But joking aside, the business model is relatively dependent on the sector. In general, people will come to us with a problem they want to solve or a space they want to explore. We will design molecules for them, and we might co-evolve and develop assets. So we kind of jointly own the assets to share risk and also share upside.
We’re also making a tool that tells you if a molecule is accessible or not, and people will pay to access that. People pay for the software access. This is computable—can I Google it? Can I Chemify it? That’s cool. People pay us for molecules, and people pay us to co-discover assets for them. I’d like Chemify to make a new perfume—there’s money in that. Some do drug discovery—there’s money in that. Maybe make a new OLED pigment—there’s money in that, and so on.
Wherever chemistry and functional molecules are needed, Chemify will be there generating new IP. In the end, our business model will be a combination of annually recurring revenue based upon making molecules, but also IP generation. Let’s say we invent a new catalyst for a particular process—we’ll do a deal where we get upside from that. The companies are quite amenable to that because they typically choose problems they wouldn’t be able to solve. It’s only win-win. If they bring a problem they don’t know how to solve to Chemify and we solve it and save them a lot of money, we win together, which is an awesome business model.
Jim: And certainly in pharma, that’s become the model essentially. The big companies can’t do shit basically, right? And they encourage small companies and then they either buy them or license from them. So that fits reasonably well with pharma. I don’t know about material science.
Lee: To be honest with you—I mean, I’m not a big pharma defender, but some of the pharma companies we’ve been working with are amazing. They can do a lot of stuff. One of the things that I think is really interesting is the chemists that work at some of these pharma companies are incredibly well versed in the molecular design landscape. Really, what we’re trying to do is give them access to chemical space faster, and then they will be able to generate assets. It’s not that companies can’t do anything. It’s that when you’re a big company and you’ve IPO’d, the stock market cycle prevents long-term investment, and molecules take a long time. What Chemify might be able to do is drop down that cycle time much faster. And if we can do that, you’ll find that they become much more productive. But that would be amazing.
Lee: To be honest with you—I’m not a big pharma defender, but some of the pharma companies we’ve been working with are amazing. They can do a lot of the stuff. And one of the things that I think is really interesting is the chemists that work at some of these pharma companies are incredibly well versed in the molecular design landscape. Really, what we’re trying to do is give them access to chemical space faster, and then they will be able to generate assets. It’s not that companies can’t do anything. It’s that when you’re a big company and you’ve IPO’d, the stock market cycle prevents long-term investment, and molecules take a long time. What Chemify might be able to do is drop down that cycle time much faster. And if we can do that, you’ll find that they become much more productive. But that would be amazing.
Jim: Yeah. We may see a reverse because it used to be pharma companies mostly developed their own drugs, but they don’t anymore. And this could reverse that. I could see that.
Lee: I strongly believe that will be the case.
Jim: That’s actually worth noting for people out there, whatever you think about the investment implications thereof. Speaking of finance, you guys are funded by some pretty impressive venture capitalists. And you talk about relatively short time horizons. VCs have relatively short time horizons. How do you see the deep nature of your work, which will take a while to ramp up because it’s a radical change and Planck’s principle and all that—how do you see your business coupling to the financial needs of venture capitalists?
Lee: That was for me a really interesting challenge. Chemistry seems capital intensive, dirty, hard to operationalize, and there’s the lure of using GPUs and AI and all that. Maybe I was thinking we could invent a new massive GPU superintelligence chemical brain called Chemgate that would magically produce molecules from the multiverse on demand. The reality is, you need to automate a load of test tubes, and you need to basically do a lot of work and get that data.
It’s not for every venture capitalist, but I was really surprised and excited that the funders we worked with, although they are venture capitalists, understood that deep tech needed to move into stuff. Our seed funder is a European-based company called Blue Yard and they really saw that. They really want to get into deep tech. Then as we went to our Series A, the fund we got was called Triatomic and they’re really excited about what they call century-defining technology.
So they’ve got the big picture, but there’s still a reality of a fund and a fund to return. The good thing about Blue Yard is they’re an evergreen fund, so they’re not under huge pressure to go through the cycle. That was one of the reasons I chose them. And Triatomic, their fund was just starting when I came in, and there was an alignment on the century-defining technology.
I have urgency because the amount of money it’s going to cost to build all the infrastructure is going to be billions of dollars in the end, and no one is going to give billions of dollars while all the manual chemistry is being done in a low-cost economy. I had to figure out an angle to make it get to cash flow positive within a venture cycle, and that was a challenge I took because I realized if I could get cash flow positive in a venture cycle, I really was going to change the world.
Chemistry is hard. It needs to be means-tested under these circumstances. The radical success requirements needed for venture was exactly what I needed as a scientist becoming an entrepreneur and CEO to force this to reality. How many technologies die because—how many Edisons and Teslas? I was like, “Oh my gosh, I haven’t done anything very interesting, and no one thinks it’s interesting, and no one wants to fund it, and I can’t make any money.” These are the triple critical points of not interesting, too hard, and not valuable.
What I realized is that the programming language is a paradigm shift because it lets us think in the same domain. That allows us to get faster and to do chemistry at a lower cost base more automatically and invent new chemical matter faster than anyone else. By doing that very carefully, we bring in the right customers. Chemify has got twelve customers now, less than four years old. We will have twenty big customers next year and it will scale on that basis. We just had to match what the market expected to what the technology would allow, and the technology was not negotiable. It wasn’t a drop of blood or a liter of blood—it was code to molecule on a cost basis that’s reliable and not burning manual labor. That was how I justified it and why the funders we have are really excited because it is going to be as significant as electrification. That’s hell of a thing to say.
Jim: I will congratulate you. I mean, I talk to and advise a number of entrepreneurs, and most of them don’t have your level of wisdom on the appropriate funding partners. Sounds like you’ve actually thought it through, and I’ve got to say everything you said actually resonates with all of my Ruttian rules on how to think about financial partners, which most people don’t do.
Jim: I will congratulate you. I mean, I’ve talked to and advised a number of entrepreneurs, and most of them don’t have your level of wisdom on the appropriate funding partners. Sounds like you’ve actually thought it through, and I gotta say everything you said actually resonates with all of my Ruttian rules on how to think about financial partners, which most people don’t do.
Lee: The wisdom that I’m kind of playing back to you here is really the patience of all the investors and all the advisers and all the people that have seen that I’m gonna make this bloody thing happen no matter what. And then I feel like a bit like Indiana Jones trying to find the right tiles to step on in the trapped room so I don’t fall to my death on spikes.
Jim: And as you do say, an interesting and probably good thing is that the venture model is a forcing function for velocity, which fits your temperament well, I would suggest.
Lee: Yeah, I get bored easily and the world has to change, and the venture model is a great way of doing that.
Jim: Alright. Let’s kick up to a higher level now. Suppose Chemify exceeds your full expectations. It’s a $100 billion company. It’s in the middle of everything. How do you see that potentially changing the structure of the chemical and pharma industries? We talked a little bit about one obvious trend—big pharma going back to its own independent drug discovery. What other things do you see about reagents and intermediate products and generic? The number of implications here are potentially vast.
Lee: I think $100 billion is not exceeding my expectation. That’s about what I’m expecting. So let’s say a trillion, and I’m not being stupidly optimistic. Like, Amazon’s a trillion-dollar company twenty years on. Now the reason I’m mentioning this is that Amazon spends more money on research than probably the NSF does. Think about that for a second. If Chemify is able to get to be a trillion-dollar company in the next twenty years and spending $100 billion a year on chemical research, just think about the biology, the drug discovery, the diseases we’re gonna cure, the new things we’re gonna have because of that technology, nanotechnology—that’s gonna be really exciting.
Going more to the non-mundane but really important: how has it changed reagents? How has it changed the pharma process? How has it changed drug discovery, materials discovery? Well, the following things I think will happen in the short term. First of all, Chemify will basically major in making rather complex molecules for people quite quickly. Not as quickly as we want, but quite quickly. And as we build more Chemifarms, we’ll just get more and more critical mass. And the Chemifarms will be dotted around the world. Why? Because we can then have supply chain security. Not just about geopolitics, just basically about load balancing. Bit like you don’t want to have all your Internet data going into one server.
So wouldn’t it be great to distribute it around the world? So all human beings, if there are changes in requirements, chemicals, drugs, whatnot, that’ll be load balanced in real time, which is very useful. And that means that Chemify’s technology that we’re using to discover also is used to make and manufacture. And that is another where there’s gonna have to be a load of funerals because the FDA and all these places are gonna say, “Oh, no. You can’t do that.” But I think we will be able to do that. The five-and-a-quarter-inch hard disk hasn’t changed form factor for a very long time, but I think you’ve gone from like a 10-megabyte hard disk to like a 40-terabyte hard disk. So I think that the Chemifarm form factor will not change, but this number of molecules you can make validated, manufacturable will just increase.
So that will mean on the reagent side, Chemify will be able to make reagents for people on demand. Hell, I’m hoping that Elon will help me put a computer on the moon so I can make drugs on the moon.
Jim: Very cool.
Lee: So there’s, I guess, four things. There is a supply chain change. There is the ability to make molecules anywhere, Antarctica, moon. There’s the ability to basically manufacture, but here’s the most exciting one: As Chemify is able to demystify drug discovery and make better metamolecules for drugs because they’re just more exquisite because we can afford to do it, think about the fact that suddenly we will go to a true era of personal medicine. Now maybe not twenty years, maybe not thirty years, maybe fifty years, but I imagine that human beings will have their own genome readout and their own drug discovery system on demand, and it will just make the molecule you need. And it will precure you ahead of time. You basically get a warning that it looks like you might get cancer in five years, and you just basically cook up the molecule you need to solve that defect and away you go. That’s kinda grandiose, like, “Oh, we might cure everything.” But you know what? We’ve cured a lot with molecules. We’ve saved a lot of lives. We’ve done a lot of things. There’s no reason to think that it’s gonna stop.
Jim: Let’s leave it at that audacious vision. Good luck to you because it would be great for humanity. You might make a couple of shekels too in the process. And as you point out, being able to fund vast amounts of research within your company probably itself is a good thing. So I want to thank Lee Cronin for a very deep dive into what they’re doing at Chemify.
Lee: Thanks, Jim. It was a pleasure indeed.