molpigs podcast #23: Ashwin Gopinath - Merging top-down and bottom-up synthesis - AI in Research

Welcome back to the MolePigs podcast. Today our guest is Ashwin Gopinath. Also here today are Boya.

Hello.

Yorgos.

Hello.

And I'm Eric. Unfortunately, Hannah cannot join us, so you're stuck with me as host. To introduce our guest, Ashwin received his PhD in electrical engineering from Boston University, working on devices for detecting and characterizing single biomolecules. Challenges he encountered during this motivated him to switch his focus from optical physics to DNA nanotechnology. leading to a postdoc under Paul Rothemund at Caltech. After working briefly at Google X, he's now an assistant professor at MIT and has received the Robert Dirks Molecular Programming Prize for his work combining DNA nanotechnology with conventional microfabrication. Ashwin, hi.

Hi, Eric. It's fun to be here.

Yeah, excited to talk to you. Um, so I guess to start, um, you, you mentioned in your introduction on your website that your PhD was a partial success. Can you tell us more about that and how that frustration led you to DNA nanotechnology as a partial solution?

Yeah, so I started my PhD in a neurobiology lab. We were trying to understand how glia differentiates into, like, progenitor cells differentiates into either glia or neurons. But then nobody told me that I'll have to kill baby mice to do this research. So about after about a year, I switched and I started working on this kind of abstract question of what does it mean to generate random numbers? Like how random can you get and what does it mean in like optical physics and things like that? And that led me to this set of experiments wherein we were trying to do localization of light on these collective arrays where you can have specific location wherein photons can get localized. And why did we want to do that? We wanted to do that because we wanted to have light-matter interaction enhanced at that particular location. So if you have a single molecule in a particular location wherein you have photons that are trapped, then you would be able to characterize it much better. And we were doing Raman spectroscopy and things like that. The data suggested that I was able to see things as small as single molecules. I was getting Raman signals from few molecules. However, it never sat right with me because in the sense that the experiment suggested, uh, I got something, but it was done through titration of molecules. We never actually put a molecule exactly where I wanted, constructed a structure around it, and then analyzed it. It was sort of like you kept titrating down a particular molecule, eventually statistically, you might have only a few molecules in a particular field of view, and you're getting a signal from it. Therefore, it is a single molecule. Didn't sit right with me. So that took me on an adventure to Paul's lab, where we started playing around with DNA origami to what was supposed to be, in my mind, half a year of, I'll get in there, there's a paper, let's finish this, get it done, I'll have another paper and move back to optical physics. It ended up becoming, you know, a much, much longer adventure. And now I do optical— they do optical physics in my lab. We do microscopy, but we don't really push the limits on it. It's not as exciting as all the other cool things that you can do. So how did you find Paul Rothman's lab? Oh, uh, uh, how did I find Paul Rothman's lab? So My thesis committee had this individual named Franco Serena, who was a key individual in creation of NimbleGen. That was a microarray company. They did optical lithography of DNA microarray chips. And he had met Paul several years back or something like that. So when I was talking to him about the problems that I was facing with my experiments, he said, talk to Paul. And this was about 6 months before I graduated. I sent a mail to Paul and completely, as Paul puts the story, he had just got literally that day, he was actually writing a description for a postdoc position, which is like he just got a grant for doing exactly what I wanted. And here it was that I sent a mail saying, I want to do this exact thing. And that's the way he tells the story. I don't know if that is true or not. But I just cold emailed him and he basically was like, okay, fine, come hang out. And, you know, we started collaborating.

It's amazing serendipity, right?

Yeah.

So how hard was it for you to jump from optical physics to DNA nanotech? Did you find the experiments relatively easy to do? Or was it kind of learning new things for you as well?

So for me, it was interesting. Like, DNA nanotech community was interesting at that point, because I came from a field, and I came from a lab that was— that had a very different philosophy, wherein it was more Publish, publish, publish, you know, minimum publishing unit and just basically push it out because the community around like optical physics and plasmonics at that time had that kind of attitude, collective attitude. The thing that was interesting to me was the DNA nanotech community, at least at the time that I started like 10, 11 years ago, was still relatively new wherein the communications between the different labs and PI were much more open. Like I went to, there was this multiple university large grant that Caltech was part of, which was called MPP, Molecular Programming Project, first molecular programming project. And I met all the individuals. And it, I mean, that's what kind of attracted me to their GN and nanotech more than necessarily the technology. The fact that it was an open field. And nobody in the field other than probably Paul and Eric and Bernie and a few others could be called as veterans of the field. The field was very ill-defined. It was sort of like a ragtag group of different PIs trying to play with this new form of fabrication tool. And that That made it much more exciting, and it also made it much more interesting. So it was not very difficult because nobody, no rules were set at that time. And and it's no long. It's it's very unfortunate that it is no. I don't get that feeling anymore in the community because it's sort of like it's grown, and there are like rules and you know how things are. done is sort of defined. At like, at that point, origami was still relatively new. It was like, there were, there were pieces that were missing. What could you use it for? If you want to attach a nanoparticle, how can you stabilize a nanoparticle? What do you do? And so that created a very, very interesting environment that made it very easy to transition in because nobody was a veteran of the field. I was actually going to follow on and ask, like, What is, or what do you mean by DNA origami 2.0? Oh, okay. So DNA origami 2.0. So I loved origami. I mean, origami is awesome. But the thing that bothered me about DNA origami always is the fact that it was DNA. And DNA from a material perspective had its limitations. You couldn't arbitrarily change the environmental condition and keep its properties. So that is to say, Uh, if I take a DNA origami, make it into some kind of 6-helix bundle or anything, and I wanna, I wanna now move it into an organic media, uh, like I wanna put it in hexane and grow something on top of it, or if I wanna, uh, let's say grow some 3,5 material on top of it, or I wanna use it, uh, uh, um, as let's say some kind of, uh, tagging agent, like let's say if I wanna make an MRI contrast agent out of that. How do I actually do all of these things? And so from a material point of view, it had limitations. So what we were trying to do, it was sort of a tongue-in-cheek title as well. It was sort of like keeping all the properties of DNA origami, that is to say, the organization of molecules, the relative distribution of molecules, the shape, the size, all of those things, but be able to completely convert into a different material so that the DNA, at least the chassis, the core body, doesn't look like DNA anymore to the outside world. So the main advantage was it's sort of building on William and Thorsten's work independently. They did the polyplexing work. And the polyplexing work allows you to make it stable in biological settings wherein like enzymatic degradation and all of that goes down. But also it makes it much more stable in organic media where material synthesis works very well. And while the enzymatic degradation is reduced in with polyplexing, it's not completely eliminated. But if you start growing silica on top of it, No enzymes are degrading silica, not anytime soon. It gets sequestered back into, uh, like it gets cleaned up, uh, by the lymphatic system, but, uh, it's not gonna get degraded. So that was, that was where we wanted to go. So we wanted to open that entire— you, you, we didn't want to be limited by the material limitations of DNA anymore.

Yeah, and I think it was Oleg Geng that showed that you could silicize a DNA crystal and then put it under like 1 gigapascal of pressure and it would stay intact, right? So does your lab also work with the silicization, or do you guys have other interesting material modifications to make to a DNA structure?

The short answer is we'll go where— we have a vague direction that we want to go. Like, my area, the thing that I want to try to solve at this point, is I want to see how finely one can actually measure molecules in all sense of that term. So that is to say, how sensitive can I make it? How broadly can I measure it? Can I measure, let's say, 10,000, 15,000 molecules at the same time? You know, what is the maximum dynamic range that I can actually get there, as well as can I measure all of these things at the same time temporally? Because eventually where we want to get to is being able to measure all the relevant molecules that are there in a biological system in real time. So how does, let's say, rather than measuring glucose, can I measure insulin? Can I measure, let's say, if you are going to take a chemotherapy drug, can I measure the metabolism of the drug? Can I, like, if you are going to take opioids, can I actually measure how the opioids are being metabolized in your body? Or, you know, at what point can I look at the patterns of proteins and small molecules in your body and be able to predictively say whether you have possibility of a particular disease manifesting? All of these things. So that's where we want to go broadly. But the technology for that doesn't exist. And so we are at that building up the technology layer. And DNA— allows me— at least it's a, it's an interesting enough material, or it's a material that's compatible enough with all the different players in this problem that we can create tools. But if we find a better material and better solutions, I'll move on to it in a heartbeat. So it's like a means to an end. So is this the technology that leads to you, um, having your own company? Can you tell us the story of that? Oh, okay. So Biometrics came about, uh, it was, uh, since I published my paper, uh, the, the Van Gogh paper in like, I think, 2016, uh, me and Paul have been like thinking about, or I've been trying to push Paul to, uh, you know, jump on board and like start a company around that technology. And that never happened for many reasons. I mean, we thought about it and we, you know, the time was not right and things like that. Then I started my lab at MIT. And what happened was, literally, I got the keys to my lab on the same day that the lockdown started. Like, I have an email wherein I have like 2 sets of email wherein like 15 minutes apart, here is your key, your lab is renovated. 15 minutes later, oh no, you can't get into the lab because it's locked down. So sort of like frustrated by that, I gave a call to like a friend of ours, And then we were like, okay, it's as good as any. We are trapped in the house, we'll just start. We've been thinking about starting a company around this, let's start it. And it was sort of trying to answer one part of the question. It was sort of like, how sensitively can you measure and how broadly can you measure protein concentrations? And so we wanted to take— we took single molecule ELISA, like sandwich ELISAs, and refactored it onto these grids that we form by DNA origami. So every single spot behaves like a sandwich ELISA. And if it captures a protein, it lights up. And then, you know, there are different tricks you play to basically do multiplexing and stuff like that. But But that's the story of how that happened.

Guess the pandemic wasn't all bad, eh? So when you were running Palomatrix, how much of your time was spent running your lab at MIT and how much of it was spent interfacing with this company, which I believe was based in San Diego, right? And were you basically having to fly back and forth a lot?

Yeah, so I had to fly back and forth a lot. But so pandemic helped in some ways because we went remote and I was remote anyway because my wife is based in San Francisco. So, and I had like, we set up base in San Francisco anyway, during my Google time. And so it was what it was. I mean, it like pandemic sort of allowed us to explore new ways of collaboration and new ways of working. So it's interesting. And to answer your question of time commitments, It's sort of, I am very fortunate, like I said, to have extraordinary students and postdocs. So sort of like, if I go into the lab, if I touch something, I often end up breaking stuff. So the truth that most PIs don't love to say or admit is that once you become a PI, you are a glorified editor. And you like to come up with like ideas on a whiteboard, but your lieutenants and your soldiers have to be awesome. And I've been fortunate enough to have some really, really amazing folks so, uh, that they make my life a whole lot easier, allowing me to, you know, explore these other things like starting companies and stuff like that.

Can you say a little bit about whether or not there's any difference in management style between managing an academic lab versus managing a team at a corporate environment?

So the thing that nobody tells you, it's sort of true even when you look back at it, even in academic environment, is the productivity of a lab or a productivity of an organization has less to do with the technology that they are working on and more to do with the interpersonal relationships and the culture that you bring to the table. And one thing that definitely is true in a startup is that it's much more mission-driven, wherein you have to get certain results done. And there is— and you have to ultimately create a product that somebody is going to buy. And it is not vague. In terms of you have to get some result that might get published, and then you have to massage the results so you make sure that you get published in a big enough journal. Uh, because so running a company or productizing something is a little bit easier, it's all— but it's also actually much more difficult in practice because ultimately, you know, you can say I have created the most beautiful technology, but if nobody is willing to buy it, it means nothing. So, so you can't, like, you can't, you know, massage the graphs and like interpretation and like, and put a, you know, interesting introduction and interesting conclusion and push it into nature science, because that can be done with results, you know, recontextualize it. But with products, if you create something, if nobody's gonna buy it, you can stand there and say, I have created the perfect technology, you ought to love it. won't happen. So it's a little bit harder, but it's also a little bit easier because you don't have these kinds of distractions. It's much more easier. It's like, it's, you can be very focused and you can have a team, all of who are working on a singular mission. It's not everybody trying to get their own first authorship paper, their own, you know, sub-contribution to this larger project and things like that. So, so that's a big difference. And it's actually a very welcoming difference. So what's the story of like your recent sale of your company to SomaLogic? So multiple reasons, right? So there's this really nice video on what makes a company or a startup a success. I think it was by the founder of Idealabs. And they did this multi-parametric analysis to kind of ask the question, is it the team that makes us a success? company successful? Is it the technology? Is it the money? And all of these things. And turns out, the singular predictor of success of a particular company is the timing. And the timing of, you know, getting into the market or basically forming the stuff. So while I can claim again that, you know, our technology was singularly awesome, the time that we started the company, was also great. But then we were towards the tail end of the biotech bubble, in some sense. So while I can say that if we had struggled on, we would have eventually been successful, it was sort of a cost-benefit analysis. You know, it's sort of like, is it going to take 10 years to become successful? And there is this period wherein we saw very clearly that the economic, like the biotech bubble was bursting. And we were like, okay, we have 2 choices, either, you know, let go of it, or, you know, struggle through several years of fundraise and, you know, and getting things to go. So, and we made a choice. And it seemed like, you know, at the time, and even now, SomaLogic had certain assets, like affinity binders, which Palmetrix could use. And is your involvement also changed after selling to SomaLogic? Significantly, significantly. I am sort of like working my way out of that. So there are still relevant things that, I mean, when you start something, but I mean, I look at it as, you know, sort of like Palmetrix was a certain arc of my life. Was a certain story. That story sort of ended. And it's sort of like it's a project has ended, it's handed off to somebody else. And they will make what they make out of it. There is a larger mission that, you know, I need to solve. I mean, that solves, that solves some part of the large mission that I had, or like, or that, that my lab still has, which is sort of, you know, how broadly can you actually measure, but it doesn't solve many of the other problems that, that Labo is a technology that solves a particular problem a particular way, and you know they are productizing it and they'll they'll make it what they make it.

So do you basically view your mission as a scientist to kind of seed ideas through the world and build up a series of technologies that can solve a series of problems rather than it being your particular technology?

No no no no no no no no no no! I I the the singular thing that I want to actually bring about is. I don't think people view healthcare with the seriousness that it deserves. Like, in the sense that I, I think, uh, and, and, and that is not to say I'm not, I'm not trying to kind of poo-poo, uh, doctors and researchers or anything like that. That's not what I'm trying to say. What I'm trying to say is, uh, like, we have technologies that are, you know, I would, I would think about it as one or 2 miracles away. That when, when I say miracle, I mean like, I, I look at any grad student finishing their PhD with a good set of thesis as a miracle. So I think about it as, uh, uh, you know, so if you have like one or 2 good PhDs away from some really awesome technologies, so being able to view your— so if you look at your body, or any kind of biological system, as truly a connected complex system, then if you can take a snapshot of all the constituents there, and if you have multiple instances of it with phenotypical information, you will be able to reconstruct the phenotypical information purely from the system information. So that is to say, if I take all the proteins and small molecules in your body and your genome, and if I were to take a snapshot of it, it's like at, you know, 9:30 AM in California, my, you know, 7,000 or 10,000 protein values is some value. This is my genome. And this is my set of small molecules in my body. Now, if I— it's— I posit that that information tells you everything about your health at that particular point. The question is, how do you interpret it? If you need to interpret it, then you need to have similar to— you need to have large data where you have collected that raw information and phenotypical information, do correlational studies similar to how you did GWAS, and to be able to interpret the patterns. For that, you need to develop technology to be able to measure everything cheap enough so that you can actually measure this. And once you interpret it, how do you actually come up with ways to get people to use it and all of these things? So I want to bring about that mode of healthcare. That's my mission. And it just so happens that nobody is building— if that technology existed, I would gladly go use it and try to build something out of that. But if that doesn't, and that doesn't exist, so I have to build it. That's it. And part of the reason why I'm getting to this is because, like, my wife was diagnosed with, like, leukemia, back when we both were, like, when I was a postdoc, and she was a grad student at Caltech. And that fundamentally sort of changed the way in which we were thinking about sort of healthcare in general. So I want to live in that world. And the technology doesn't exist for that. So I'm building it. If the technology happens, then the next question is, how do I get people to use it?

So when you say people like aren't taking medicine seriously, do you mean kind of they're too siloed in like, one biomarker, one phenotypic outcome, whereas you think that the actual answer is to be much more holistic and much more integrative in your medicine approach such that you take into account the system, how the system is behaving as a whole. Do I kind of have that idea correct?

Correct, correct, correct. I mean, like, for instance, if your computer is basically slowed down, you're not going to ask there, okay, what is the one you know, particular thing that went wrong or one particular software that basically is causing the slowdown. It might be the case, but more often than not, it's a series of things that basically knock on to each other that causes any kind of problems. So, I mean, that's, that's not a, that's not a good enough example. Maybe. I mean, it's hard to come up with complex system examples of complex systems wherein it's viewed holistically. I mean, like, where— or wherein it is not viewed holistically. Like, when you have car trouble, you know, you— I don't know. I mean, I mean, maybe, maybe these are not the right examples. It's like, I'm having a hard time imagining what would be a right example to bring up there. But, and again, this is sort of intuition, and which is sort of supported by emerging papers, that you need to view biological systems in its full complexity, and we don't have tools to even view it yet.

So that kind of makes for an interesting segue into what you've been posting a lot about on your blog recently, which is artificial intelligence. Do you kind of view artificial intelligence as the way we're going to handle this integration since it can handle so much larger data? Or are you more focused on artificial intelligence and how it can interface with humans?

Oh, so the whole adventure with the AI sort of began because there were too many papers. So there are too many papers that are coming out, and I can't keep track of it. So I started coding up a little tool for myself, wherein I just take the entire bibliography, and it does a semantic search on the entire bibliography. And I started— I mean, it was a really, really badly written code for myself. And then I found a bunch of students. And it turns out, we built a tiny little system wherein you can just put in any arbitrary books or any kind of YouTube video or any kind of things, and you can start interacting with it. So it evolved into something a little bit bigger. And it's just sort of snowballing. It's like taking its own life right now. The AI— and we are, we are exploring, I mean, there is a paper that should be coming out in the next couple of days, wherein we've been playing with how do you actually get self-reflection in AI. And we are trying to do that for— we are designing that for protein design. So that is to say, if you give a particular protein target, it semantically tries to create the structure. And then when it fails, or it doesn't hit the target, it asks itself, you know, why did I fail? and come up with hypothesis in natural language, and it reruns it. So sort of like, how can you actually come up with that system? And what does that system look like? And does it actually perform better? And so we have like the core agent, the core architecture should be out in the next couple of days, but we are trying to plug that into molecular design. So much more precisely, now the question is, if you are trying to design an origami, What is your design process? Or if you're trying to design a molecule, you will go, you have certain knowledge, you will try to design something and you have some measure, some heuristics by which you say, did I actually design it or not? And if you didn't design it, you're going to stare at your design and you're going to come up with hypotheses as to why you failed. And you will try it again. Now, can I systematize this entire process and let it loose? So, uh, like, so that's, that's where we are going. So the first step of that, that should be out in the next few days. But then we are applying that to molecular design, which will take a little longer. So it seems that your research philosophy is very practical-focused. So I wonder, uh, how do you develop this kind of research philosophy as describing your website? Like, do you— did you um, go through some trial phases on theory and realize that, oh, you didn't enjoy it? Or did you have a passion about practical and experimental investigation from early on? So, so partly being around, being in Caltech and being in Paul and Eric's lab radically changed the way in which I thought about science. I think more than anything else, I feel like Caltech. I've been associated with different organizations and universities, but I think more than anything else, I think those 2 labs kind of define what I think of as being scientists, uh, like thinking about it in a certain way and not like wherein you don't think about it strategically from an academic point of view. What I mean by that is Many times people think, and I, and I, and I hate that academic community has become this, wherein as soon as an idea comes in your mind, you're asking the question, can I get a paper out of it? That's the wrong question to ask. The wrong— the question is, what is a problem that you're trying to solve? And is your work actually moving you towards that problem solution? And if you get— if you do good work in that, and you are successful in that, Good papers will come, as opposed to asking the question, let's kind of do something and like make sure that we have papers. Okay, and so so so those set of those labs sort of made made me starkly realize that it's possible to kind of maintain that purity in some sense of doing science and and and still get away with it. It's hard, and it also requires one to sort of. Throw caution to the wind to a certain extent, because you're doing things that are like way off the left field, wherein like you're not trying to constantly publish, you're, you're not constantly trying to go to conferences to give talks, you're sitting there and you're like, this is the problem that we are trying to solve. This is the way in which I have a particular vision on these are the steps that I want to do. And these are the problems that I need to solve. And you go after that one and one after the other. And if you get papers out of it, great. If you don't get papers out of it, we'll deal with it. I mean, publications has its own problems in academia, but, but that's, that's, that, that's— I'm not even gonna go start with that.

And how do you get that philosophy past like the tenure committee who are looking for, you know, publication metrics and conference talks, etc.?

Uh, how do I put this lightly? Not caring so much about it. It's as simple as that. It's sort of like, uh, it's sort of like you have a, you have a choice. I mean, I, I asked myself this question a while back, and maybe, I mean, it'll, it'll, it'll come back to bite me in the ass at some point, wherein if I might not get tenure, but it is what it is. Uh, but the question is, and then I'm okay with it, but it's just sort of, uh, it's just sort of like, you know, what do I want to do? Do I want to spend the next 6 years or 7 years of my life basically constantly stressing out and writing papers and chasing things and doing stuff chasing problems or chasing— playing a game whose rules are set outside me, or do I want to actually build something that I want to build? And, you know, at the end of the day, you know, you have a choice. So, and I'm choosing to, uh, you know, hold steadfast, and we'll see where it goes. We'll see where it goes. It's, it's sort of worked out for me so far. Maybe the luck will run out at some point. If that's the case, there's something else to do. So, you know, I don't want to play the game by somebody else's rules. That sounds very free. Does that also explain like why you— I saw that on your research website, you have very detailed description of the project that you're interested in. Is that also partly explained with like style of open research? Yes. Yep. I mean, I've been trying to tell my postdocs and grad students you know, as soon as you have results, just put it out there. It's okay. I mean, it's sort of like, you know, sitting and worrying about it is just sort of like not going to help. I mean, uh, you will— you might get it into Nature, you might get into Science, it might— I mean, I think, I think, like I said, what I want is technologies that allows me to answer certain problems, like solve certain problems. It doesn't matter whether it comes out of my lab or it comes out of some other lab. I think the problem needs to be solved. And, and, you know, I'm not going to try to claim that, you know, the problems that I'm trying to solve is as big as like climate change or anything like that. But it like those are places wherein it actually matters, like publications and the pace at which publication is done. Like it hurts it. Because if, if climate change really needs to be solved, and these kinds of big issues need to be solved, then you want to put out— you want to collectively work on solving the problems. Like, everybody should be like— and, and the pace of innovation will be a lot better, a lot more if you just put out your ideas. Who cares whether you're doing it or somebody else is doing it? Like, if somebody else like takes some of these ideas and actually works on it and develops something, great, it solved my problem. I don't need to solve it. As long as I can reproduce it, I'll build on it.

And so I've noticed from looking at like your Google Scholar that recently there's been a ton of patents out of your lab, and I see— I think a lot of those are Palamare genomics-related stuff that only finally got granted. But is that kind of your way of saying, okay, we did this, as in like the grand vision of single molecule detection?

Not, not really, right? Like, so, so patents in a startup or like in a commercial setting, you know, is, is purely from a strategic point of view to allow you to work on certain things. So that's sort of the reason why we put it out there. And there are a bunch of other patents that will come out soon from my lab itself. But those have to wait a little bit. I mean, those are, those are still sort of cooking. But patents are, unfortunately, patents are the way in which deep tech works. I mean, it's sort of, you know, we can have a, that's a, that's a longer debate as to whether it is good or bad. There's a lot of stories there as well. So just because you have a patent doesn't mean that you can work on certain areas, and just because a publication has happened, it doesn't mean that you can't still patent it. There's a whole bunch. It's a, it's, it's a tangled web.

Very true. Um, I'm kind of curious to hear a little bit more about how you took those ideas from working with Paul and Eric's lab at Caltech and then transplanted those to your own lab. Um, how does that really influence the creation of the lab culture that you are focused on, that you seem to be trying to implement as well?

So, so my lab, my job, I look at it as my job as a PI is making sure that funding is available, uh, to play with things and that the students and people in my lab Don't have to worry about that and trust my team that, you know, we will like sort of vaguely make sure that the ship is pointed due north, so to speak. I mean, if you're going due north, it's sort of like it's pointed in that direction. I just need to make sure that, you know, the steering wheel is in that direction and I put fuel— that there is fuel in the fuel tank. The rest of it is just basically up to the students. And I'm sort of like, they have the freedom to mess up. And we have a clear understanding as to what they want from the PhD. We have that discussion really early on, wherein do they want to be a PI? Do they want to go into the industry? Do they want to build something of themselves? Or if they don't know, that's also fine. But, but the, but the question is, uh, the question is the, the strategy is different because if you wanna become PI for whatever reason, uh, uh, then you need to kind of focus on putting yourself out there, trying to build on your ideas, trying to figure out how to refine those ideas, how to communicate those ideas, build relationships with PIs. You know, build relationships within the community and things like that. But if you want to start a company, or if you want to go become a research scientist or something like that, then you need to have exposure. And you need to make sure that, you know, they have— they do the internships and all of that other fun stuff, so that, you know, when they graduate, they're not— they don't think that writing papers is the skill that the industry, you know, rewards, because it doesn't. I guess following on from that, how have you, like, practically, how do you shape a graduate student who wants to go into industry, you know, for 7 years in graduate school? It's not 7 years, it needs to be quicker than that. Oh yeah, 4 or 5 maybe. Yeah, yeah, 4 or 5 is ideally, but, but 4 to 6, let's say. The thing is, If they say that they, if they say that they want to become a PI, one of the things that is essential is we need to kind of learn how to write, like learn how to write and write proposals and, you know, learn how to give good presentations in a certain way. And, you know, expand the ability to collaborate with other PIs that they are leading themselves rather than, you know, I'll always be there to kind of stand in and just, you case they falter to make sure that this is there. But, but I want them to lead it because they want to have those relationships with PIs so that when they ask for reference letters or anything like that, that there is an existing relationship. Now, alternatively, if it is someone who wants to go into, you know, startup work, you know, or, or, you know, industry, it's actually very liberating because that basically means that you don't need to pressure them to write papers, or at least like chase after Nature science papers, first of all, because it won't help. I mean, it doesn't hurt, but it won't necessarily help. Nobody's going to basically say, I'll give you a senior scientist or like some staff scientist or some position because you had a Nature paper. Very rare, very rare. So, you know, what are the, what are the things, what are the skills that are needed there? And make sure that they have breadth as well, so that they are not sort of so narrow if they're doing DNA nanotech, they're not only thinking about DNA nanotech, they're thinking about things beyond it. Like, let's say, you know, high-throughput proteomics, high-throughput genomics, or if you want to go into, let's say, drug development, how do you actually think about that? You know, whether it is internship or like through collaboratively, like getting them to go to conferences or whatever it is. But ultimately, it is sort of like, at least for me, it's to make sure that I have that community conversation with my students and even postdocs early on and continue to have those conversations. I mean, because otherwise it's just sort of you're training them to become another PI. That's the typical way in which most, uh, graduate school goes through. I mean, and, and, and that ultimately ends up hurting you.

So as we get close to the end of time here, I always like to ask a like kind of forward-looking question. Um, and so you've You've been through a lot of different research environments. You've worked in academic labs, you've worked for Google, like this big company, you've led a small company, and now you're really focused on how AI can change our interaction with science. How do you see science changing as this technology develops and as different ways of leading labs becomes More common.

So one thing, one thing that I am sort of looking forward, and I'm getting, I'm going to get like a perverse pleasure from seeing this happen, is just sort of like the way in which generative AI is going to basically destroy grant writing and academic publications. It is going to, I mean, it is only a matter of time. I mean, because, because I can churn out papers if I want, you know, extraordinarily fast. And if I'm using that for like grant writing, I can, I can put out a grant pretty much every day if I want. And if I have a core grant kind of written, I can make themes and variations of that constantly. So just like there was this very famous sci-fi comic that had to turn down, turn off its submissions because of the deluge of submissions that they ended up getting. I see that happening in NSF, NIH, DOD, every granting agency. If they open it up, they're going to have like, you know, tons of submissions. We are just beginning of what we are going to see. It's the same thing that is going to happen with academic publications as well. Like, how can you stop somebody from, you know, you don't stop somebody from doing using spellcheck. It's as simple as spellcheck to generate all like themes and variations of the same idea. And, and you can't keep up with it. It was all the strain of academic publications was already being felt by academia. This is just going to make it worse. It's going to be interesting to watch this change. However, there is also opportunity for interesting stuff to happen. Like, for instance, you can have completely autonomous learning. Like, you don't necessarily need a lecturer or like a professor to teach you a lot of things. So that makes a lot of That makes life a lot easier. AI, you know, all the fun stuff that can come out of, you know, throwing this unstructured data into some kind of singular database and looking for unsupervised learning on that corpus, that would be interesting. That's coming. But, uh, we just have to see when that happens. Sorry, I have a little puppy here. That's why.

Dogs are always welcome on the podcast. Um, so on that note, I assume based on your answer that you've been using ChatGPT in your, like, writing day-to-day recently. How has that experience been?

So I don't use— I don't use ChatGPT often in my writing because it's just sort of, uh, the way in which I— the way in which I use ChatGPT is not ChatGPT necessarily. I have like my own little scripts for doing some of these things. What I do is I have an initial draft of ideas that I basically write down that where and when literally like bullet points. And I have a corpus of other data that basically I have, I appreciate and basically like the design of like the style of. And I basically what I do is I ask my tools to Write rewrite my bullet points in the style that I want. So that makes it like out pops. So I can basically it makes me uncomfortable. It makes me very uncomfortable to actually even state it, but it makes me very uncomfortable too because because it's no longer my voice. It's sort of like amalgamation of voices that I like, the styles that I like. as well as like the points that I want basically expressed in that way. So, but it's coming. Have you ever thought about training a model on a GPT model on a corpus of your own grants and just churning out grant after grant? That's exactly— that's exactly like, I know it can be done. Like, it's not just my grants. It's like a whole bunch of other grants that I have basically access to where you're starting a— when you're starting an academic lab, you have your collaborators and your colleagues sort of like share the proposals with you. And I've not explicitly asked anybody's permission for it. But they have shared it with me. So I've basically loaded it into my program. And it just basically pops out the style. It's like an amalgamation of that. Which is what like I've considered basically making this open source or like putting it out in the world. But it's just sort of like, it adds to the noise. And if somebody else is going to do this anyway, I mean, I'm not going to do as good a job as somebody who's a, you know, trained software engineer. So I'm using it on my own, like, like, these are just simple Google Doc scripts that you can just run. So it runs within Google Doc, and it knows a database of grants from which it has learned the style. So I just need to pick a set of bullet points and say, make a grant from it. It does— there is some nonsense written there, but still it's, it's within striking distance. So, and this, I, I am, I am, I am 99% certain, and I'm— I, I, I know few people who are also building this out, uh, uh, uh, this is going to be common in the next year. Is like a year from now, if we are chatting You— everyone will be using this.

Guess we'd better learn fast. Thank you so much for joining us, Ashwin. Stay tuned to our newsletter or Slack channel for details on our next podcast episode, and thank you for listening.