Thomas Miller, Theoretical Chemist and Biotech CEO

What are the translational possibilities of theoretical chemistry in a new world of artificial intelligence? In the middle of the pandemic, Tom Miller decided to pursue this question in a dramatic way. After a career in academia and a tenured professorship at Caltech, Miller concluded that a major career change was necessary. As cofounder and CEO of Entos AI, Miller entered a new world of entrepreneurs and venture capitalists, all on the basis that the science was so compelling, and the medical need so great that success required a full time commitment.

After his undergraduate degree at Texas A&M, Miller completed his thesis research at Oxford University during a time when advances in computation were revolutionizing the field. He focused on statistical path-integral methods, and their use in low-frequency motions of molecules. For his postdoc, Miller worked with Bill Miller and David Chandler in pushing the methods he had developed at Oxford to apply them to ever more complex systems. Miller then joined the Caltech faculty in 2008. In the conversations below, Miller speaks eloquently of Caltech's culture of supporting its junior faculty, and the positive pressure this creates to push the boundaries of discovery. Following achieving tenure, Miller significantly expanded his research agenda, and his group worked on projects ranging from catalysis, energy efficiency, and soft matter dynamics.

By 2018, advances in machine learning began to convince Miller of new and exciting possibilities in both the kinds of questions that could be investigated and the chances of translating basic science to societal benefit. In making the decision to found Entos AI, Miller muses on the value of demonstrating to others just how strongly he believes in its potential. Only three years from its founding, Entos has grown significanty and it is pioneering the use of artificial intelligence. The company's progress so far portends to exciting developments in molecular discovery and drug development.

DAVID ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Wednesday, June 1st, 2022. I'm very happy to be here with Dr. Thomas Miller. Tom, it's great to be with you. Thank you so much for joining me today.

THOMAS MILLER: Thank you, David. This is a pleasure.

ZIERLER: Tom, to start, would you please tell me your current title and institutional affiliation?

MILLER: I am CEO and Cofounder of Entos, Inc. I am still a Visiting Associate at Caltech as well.

ZIERLER: Now, let's just start first with the name, Entos. What does that mean? What does it convey?

MILLER: It's Greek for "inside" and it reflects one of the original technologies that spurred our translation from academic to industrially oriented research. We can talk about it in detail but the notion was that if you're going to use computers to predict chemical quantities, chemistry tends to be local, around reactions. You have a little bit of a chemical system that really is doing the business but it happens in a liquid, or it happens in a protein. It happens in some big environment. You would like to have methodologies, computational tools that will accurately describe that important region while embedding it in a reasonable but less accurate and more affordable description of the environment; thereby embedding a high quality within a lower quality computation description. That was one of our main initial technologies. It originally led to a program, and then we took the name of the program for the name of the company.

From Basic Science to Translation

ZIERLER: Tom, before formalizing the company itself, what were some of the intellectual kernels coming even from fundamental research that made the idea of Entos feasible at that time?

MILLER: On the one hand, in my research activities, I try to embrace the theme that there should be ways to incorporate rigor and higher accuracy from fundamental physics into the description of complex systems. You don't just want to do high accuracy on very little systems, and low accuracy on big systems. There should be ways to take the good ideas from one place, and to learn how to scale that and to translate that to more complex and interesting systems. That was something that was, I think, a prevailing theme. It is one kernel that led to the translational research that led from my group to a start-up. I think another kernel that is less scientific and maybe more philosophical is around problem selection. A lot of times in one mode of operation, you will say, "I'm going to work on the problems that are particularly interesting to this audience, that are particularly important to my field." I think that allowing myself and allowing my team to think about different definitions of what constitutes an important problem, and different scales of what is an important problem, opened the door for a more translational direction in my work as well, and led from Caltech to a start-up as well.

ZIERLER: Tom, just as a snapshot in time, I know Entos is still in growth mode. How small was it when it first started? How big is it now? What are the plans for future growth?

MILLER: It was Fred Manby, my long-time collaborator, and me as of, I would say, just before April 1st of 2020. That's when we basically launched it. Now, it's just basically two years later, I think we're at 65 now. It's grown. About half of those are software-, algorithms-, computationally oriented people. Then the other half are drug discovery scientists, chemists, biologists.

ZIERLER: Is Entos involved in basic science or is it exclusively working in a translational capacity?

MILLER: I would say that we innovate, and we take pride in innovating and doing things that are novel and that would count as publishable basic science. But unambiguously [laugh], our value proposition is making better drugs. If it's not in line with that, it's hard to justify.

ZIERLER: What are some of the connecting threads in terms of the different kinds of products you want to bring to market? What connects all the things that Entos does?

MILLER: The core thing that Entos does is that it combines AI or machine learning in a very tight coupling with high throughput experimentation. It basically means automated experimental activities, including both the synthesis of molecules and the assaying, testing of those molecules in biochemical and in cellular assets. The whole point of what we're building is a tight coupling between making good predictions of the many properties that a drug molecule has to satisfy, synthesizing those molecules, testing those, and then generating data for those important properties that will drive the next generation of molecules, and doing that at the scale of thousands of molecules per week. It's a lot of exploration very quickly of chemical space, driven and guided by machine learning, and enabled by those automated strategies for making and testing the molecules. Really, at the heart of that is the ability to do the predictions, the ability to learn from that data, and the infrastructure to connect all of those technologies together in a way that takes advantages of those different capabilities, and pursue the development of better drugs. We specifically apply that to two internal drug discovery programs. We're looking to expand that but, currently, we have two programs, both in the area of oncology.

ZIERLER: Is it cancer that was sort of the initial driver for you to get into this field?

MILLER: I think the answer is no. Really, the first year that Entos came out of Caltech, throughout the calendar year of 2020, we were really interested in the general ability of using machine learning to make better chemical predictions across many industries. We work with Dow. We work with Procter & Gamble. We've worked with Zymergen. We've worked with pharma companies. We've worked with a lot of different materials companies, Toyota. We've worked with a lot of different companies for a lot of different chemical problems, polymers, drugs, catalysis, all these different things, because it was a very general software tool. In that first year, we showed that the technology worked. We put it in the hands of not just ourselves, but we put it in the hands of those customers, and they were convinced it worked, and they spent money on it. That validated the technology, and this enabled us to make a very strong case that, well, this software works, and it does something that adds value, and gives you an advantage to design better molecules. We decided to turn that inwards. Let us combine that with the right experimental counterpart. Let us point that at the right important problem, and then let's really take advantage of the value and the impact of what we're doing, as opposed to just giving it to other people to use in their own applications. You stop a service role and start playing a product role, which is what we wanted to do. That logic leads you very quickly to drug discovery. It's a very natural fit for these technologies. Small molecule drug discovery, it's an area of huge need. There are, of course, many areas of disease. Then you start asking yourself, OK, where are the areas of biggest need? Where is the fastest area in order to get to clinic? Different questions like that. Oncology is among the natural choices for that. Then you have to focus, so you focus on that.

ZIERLER: Why oncology? What is it about cancer research that makes oncology a natural outgrowth of these questions?

MILLER: There were many. It's just a sad reality that it's an incredibly severe disease. It's an incredibly fast, quickly advancing disease. Many people are afflicted by it. There's many varieties of it. It is the combination of those things that means that if you have the ability to design a new drug, there's a way to find a niche in that space pretty easily, and to have a relatively fast timescale to advance that to the point where it's in human trials. We're not even at calendar year one of our drug discovery effort, and we're already into animal studies. We didn't have labs in July. We didn't have any chemists until September. All we were doing was just a bunch of software people, a wonderful team. What we've been able to do in that intervening time is build the labs, design the molecules, identify the targets, get initial hits, optimize the leads through biochemical cellular and now into in vivo assays, with the goal of getting this in the clinic in the next, I guess, 18 months. That's a pretty fast pathway to get to the point where you're actually starting to see a human impact with that work, and that's exciting. That wouldn't be the same in all disease areas. That wouldn't be the same in all product development areas. Those can involve much longer timescales.

ZIERLER: Tom, launching in April 2020, what were some of the obvious challenges and possible opportunities of doing all of this in the midst of the pandemic?

MILLER: It had difficulties and joys, I would say. I think that some of the downsides are the fact that you have to build a team of people that have never met each other in person. But the upsides were manyfold, I would say. Take recruiting: Relocating people was no longer as much of an issue, so we got talent by recruiting people all over. The fact that we initially focused on software as opposed to laboratory activities worked really well in that mode. We basically didn't create laboratory space and then, of course, running experiments until a year in when the pandemic was much more of a known quantity, I would say. Other things: fundraising can be brutal in terms of travel, in terms of going to New York or you go to the Bay Area and go into the financial hubs. I didn't have to travel at all [laugh] so it was great. I did it all in my pajamas. That made it all much easier than otherwise would've been the case. The pandemic definitely had its challenges, but you have to keep an eye on culture, and you have to keep an eye on forming cohesion in the company.

Academic Expertise in a Business Setting

ZIERLER: Tom, to give a sense of your own area of specialty, what is your expertise in relation to what Entos does overall, and who are the other key people that do things that are really not in your sphere?

MILLER: That has definitely been interesting. I spent my entire life as a theoretical chemist, and my entire professional career as an academic, so no industry experience. Never really worked in a company. No drug discovery experience. My expertise is that I'm a good scientist, and I'm an expert in the areas of computation and, to some extent, in machine learning as well. We've contributed to some advances in that area that we're very proud of. But in terms of my current role at Entos, huge swathes of the company do things that I have had to pick up and learn in the last three, six, nine, twelve months, and that's a very interesting thing. Our executive team now includes—we haven't announced it but we've hired a CSO, a chief scientific officer with 25 years of industry experience, and 2 drugs to market, and 15 to clinic. We've hired heads of chemistry, heads of biology, heads of analytical technologies, heads of business development, heads of compound management and logistics. We have basically a team of 10 executives at the VP level or so all of whom have 20-plus years of industry experience. Then I'm the CEO, and my cofounder, Fred Manby, is the CTO. We've played a big role in establishing the direction, and working with the team to kind of point all of that expertise in a unique direction that really builds a company at the interface of technology and biopharma, and that's a unique thing. It's really kind of marshaling that vision and trying to make sure that people are doing what is needed to execute on that so that we maintain a competitive advantage; that is the real role. It's quite a bit different from being an academic in obvious ways. But in some ways, being an academic is all about the same stuff. You've got to fundraise. You've got to write grants. You've got to recruit. You've got to get graduate students. You've got to convey a research mission in an independent research program. It's all the same stuff. It seems very different in some ways, and very similar in others.

ZIERLER: Tom, was the original idea to do this start-up within the Caltech infrastructure to maintain a Caltech professorship, or you knew from the beginning that, at some point, this would become a full-time venture for you?

MILLER: I definitely kicked the can down the road in terms of my own thinking on this. Caltech has a very supportive policy of sabbatical. I'd never taken a sabbatical. I'd been at the university for 12 years—I think that's true—yeah, 2008 to 2020. I'd been a full professor since 2013. The transition really kind of came in creeps. I originally thought that I will take six months off. OK. Now, I'll bump that out to a year. OK. Now, we want to do a second year. Entos itself became more than just kind of a fun thing where we were going to take our software, and sell some licenses, to where, now, we've done a serious raise, and that carries the implication of building a different team and a different timeline. It's been a gradual process. It was not the original intent to take that leap but it did not feel like a big leap in the end.

ZIERLER: Tom, in thinking about how supportive Caltech is in entrepreneurial ventures, I wonder if you have an historical appreciation of a time at Caltech when that was not always the case, when people like Lee Hood needed to leave because that was not what you did at the university. I wonder if you had that appreciation.

MILLER: I am familiar with that, and that's a little bit field specific. Caltech celebrates its past entrepreneurs: Beckman; the founder of Xerox whose name alludes me. I think that it definitely celebrates that history. I know that it hasn't always been—there have been individuals who, yeah, there's the Lee Hood story, and there's other examples like that. I just think the academic community in general has, field-by-field, struggled in different ways or grappled in different ways with the right way to acknowledge impact in tech transfer or commercialization. There is concern that monetizing science can lead to less open science. It varies field-by-field, and software has been one of the natural areas of debate on that question.

ZIERLER: Tom, to give a sense both of the start-up culture and in basic science, when you told colleagues that you were thinking about making this career jump, did it seem normal? Did they say, "Are you crazy?" Are you giving up a tenured professorship? What was the general reaction when you started to tell people of your decision?

MILLER: It made it very easy that I was able to do it after a long time. Before I communicated it to anybody, I'd already been gone for over a year. With the pandemic, everybody was kind of gone for the first year. It's like my absence wasn't really missed, at first, because everybody was gone. Then it was like, "Wow, Tom's been gone a long time. I wonder if he's coming back." The seeds of it had already been planted. I think many people who had thought about it had already, at least, suspected that this was something that I might not return from. But that's not really your question. Your question's like "what's the response"? I feared more negative responses but from no colleague did I receive anything—this is a true statement; it's not an exaggeration or white-washing—from no colleague did I receive anything but support. A couple of people were like, "Yeah, I would not want to do that. [laugh] That does not appeal to me." [laugh] Some colleagues were like, "Man, I'm jealous." [laugh] Some colleagues were like, "I can see that if I had made this move in an earlier part of my career, some of my companies that I've cofounded would've been more successful. I find it very interesting you're doing this." Some people were like, "That seems fun."

But there were all sorts of different responses about how some people found it appealing or not personally appealing. But nobody was unsupportive in it, and that was better than I could've hoped for, and it certainly meant an awful lot. You've given your entire professional career up to that point to a university. It means a lot to you, and to leave it well means a lot to you, and to leave it on good terms and to leave it with all the friends and relations that you've cultivated for over a decade, and to not feel like you're walking away from all those friends or all those relationships—and it didn't feel that way in the end. I guess I feared it would feel much more like that, and it really didn't feel that way in the end, which was a real blessing.

ZIERLER: Did you see the trajectory of your motivations over the years increasingly veering toward translational questions that might not be best pursued within an academic environment?

MILLER: Maybe. I think so. I think for my assistant professorship years, I was an academic's academic. I was really focused on single projects, single PI contributions, in a very focused way. I think after I got tenure in 2013, even if you just look at my publication record, the number of collaborations explodes. I started thinking, oh, that's fun. I'll work on this. Oh, that's fun. I'll work on that. Seeing the tools that we had developed start to manifest in different application areas of chemistry or materials really did appeal to me. It probably did lead to a sense that, well, the next step in that is to really knuckle down—not to contribute in bits and pieces across many different fields—to try to really choose one, and to really sink your teeth into that, and to make a bigger contribution on that. I think that that is kind of the next stage of the translational progression that entered my mind.

ZIERLER: Tom, in having maybe two generations of graduate students plus postdocs over the course of your career as a professor, what kinds of things were they doing in terms of both their scientific motivations and their career motivations that might have exerted an influence on you and the decisions you made?

MILLER: I think that people who'd come through my group, many of them had gone on to become faculty members and academics. Some of them had remained in science but on the industry side, but that was less common for my group than academia, I would say. Some of them went into other areas. Some did start companies, and I did find that exciting and impressive. I think the students were a little surprised—I don't think they foresaw me taking this step as much. I feel like it was my own internal impulses that were leading it more, as opposed to the influence of the students around me. Maybe that's not correct but I think that that is the way I see it.

ZIERLER: In terms of the lines that you have to draw, keeping a firewall between your students and the business, did you see opportunity? You had all of this talent. In going full-time, is there more recruitment ability where you can take some of these graduate students, and put them in a different context that you wouldn't otherwise be able to do?

MILLER: Yeah, that's a careful line you do have to walk. One of the good things is I had many former students who became early employees in the company. It was a really wonderful thing to build a company where, I think, the first 20 employees, I had personal relationships. Not all were from my group but I had previous scientific relationships with at least 18 of those. It was a very high fraction that I was able to draw on from my academic career and the network that I had there, people that wanted to join the adventure. It's also clear that Caltech allows for industry collaborations. There were some very natural collaborations, even co-filing some IP between the company and Caltech, some natural collaborations. But you're absolutely right. You have to firewall. You have to be very clear about the nature of conflicts of interest. You have to be very protective of the student's best interest. Caltech keeps that very much in their mind, and it's my obligation to the students as well to also keep that very much in mind, and to just be very open about that interface. We always work very closely with Caltech about that, and they were very helpful in that too.

ZIERLER: Just a geography question. Is there a biotech start-up culture in San Diego? Is that why you're in San Diego?

MILLER: Yeah. San Diego's probably the third biggest US hub. The Boston Area and the Bay Area are the two largest, but San Diego is probably the third-biggest area. It was nearby. It's bigger than the LA area. Amgen and Thousand Oaks are in LA but it's much smaller than the San Diego ecosystem.

ZIERLER: What's the nature of your current affiliation with Caltech? Is it essentially a courtesy appointment?

MILLER: Yeah, I think so.

ZIERLER: Can you sit on committees? On a day-to-day basis, what would that mean to have that appointment?

MILLER: It means that there's a couple of remaining students that are finishing up in my group, and we're continuing them through. That's going to conclude over the course of this calendar year. I'm not actively serving on any other committees in any kind of way. It allows me to keep my email address and that kind of thing. But it's a pretty complete transition.

ZIERLER: Well, let's go back now to the beginning, starting with childhood. Were you always interested in science?

MILLER: Not in the way that you might always—I was always a curious kid. I always enjoyed science. I came from a family where you were expected to do well in school, and had very supportive parents. But I was never one of these kids that played with chemistry sets or anything like that. No, I liked sports. I liked to run around in the backyard. I think the first real interest that was notable was probably around 10 years old when I started to play the cello, and then I became very, very into that, actually. For the next, oh, I think until I was about 16, that was my life, I would say. I even lived away from home, and was at a music conservatory at SMU, Southern Methodist University, in Dallas. I took that very seriously. But, no, science really emerged as a passion once I ceased playing cello, in the last year and a half of my high school career is when I really kind of fell in love in chemistry, and I focused on that.

ZIERLER: What about computers? Did you have any interest in computers, or talents? Is there a backstory there?

MILLER: Yeah, I remember being in my high school classes, and you have some kind of basic programming, and there were always some kids who were just like real wizzes, and I was not one of those kids. I got very good grades. I always did very well in that. But I wasn't a singular computer guy either. My interest in computers grew when I was a freshman at Texas A&M University. I initially entered the undergraduate degree as a chemical engineer, which at Caltech [laugh] I would love. That would be amazing. At Texas A&M, it means you're basically learning how to build chemical plants, so that was not as stimulating to me at the time.

From College Station to London

ZIERLER: Much more applied at Texas A&M, you're saying?

MILLER: It was much more applied. What I really liked was the interface of chemistry and mathematics. Within a very short period of time at A&M, I initially changed to a chemistry major. Then I was invited to a math competition, and somehow, I won the math competition [laugh] so then they invited me to be a math major as well, so I dual-majored there. Then you're doing chemistry and math. But the entrée into real computing was still in that first semester at A&M, I got one of the graduate handbooks. Basically, the research programs were still described in hard copy, and they had kind of a glossy brochure of all the different research groups. I saw one of the professors was using computers applied to chemistry, and I was like, oh, that seems really cool. I started working with him—that's Mike Hall—and I worked with him all four years at A&M, and just had an amazing time.

ZIERLER: Who was it? What's the professor's name?

MILLER: Michael B. Hall. He remains a very good friend and is supportive. I saw him just this last month or two ago.

ZIERLER: What kind of research was he pursuing at that point?

MILLER: Basically, inorganic chemistry, transition metal complexes, organometallic complexes. Structurally: what do they look like? What are interesting ways in which they'll bind small molecules, hydrogen or carbon dioxide or carbon monoxide, for different catalytic processes? He worked at the interface between inorganic chemistry and computation.

ZIERLER: What could computers do? What research was made possible as a result of taking a computational approach?

MILLER: Theoretical chemistry by that stage—this is like 1998—actually wasn't the dark ages of theoretical chemistry. It had already gone through kind of the joke period where you could do four atoms, and the result was not good. It had already gone through that very immature phase, and it was already at the stage where you could credibly do complex-looking molecules, things that a synthetic chemist would be interested in, and you could start to gain real insights into the transition states that reactions would go through, the different conformations that they would adopt. You could really do pretty good calculations by that point, I would say. They were slow. They would take days. They would be less accurate than they are today. But it always felt like a pretty useful tool, and something that you could add a lot of value.

Now, I really did love that connection between mathematical equations and physical reality that computational chemistry offers. You get the Schrödinger equation. It's like an ordinary differential equation problem. You can take that, and that's math, and then you can solve that equation, and out pops a wave function, and that wave function literally is the bond that tethers two atoms together. That's amazing. That's just a miraculous thing, and to be able to not just see that in a textbook but then to say, OK, now that I understand that connection, I can use a computer to help me solve that, and I can really anticipate new molecules with new properties. That, I just loved, and that drove my enthusiasm.

ZIERLER: The satisfaction in getting those results computationally, how do you compare that with work in a wet lab?

MILLER: I never got the same level of gratification in a wet lab. A lot of computational people will tell stories about how they have two left feet in the lab, and that sort of thing. I was not a talented laboratory chemist. I wasn't abysmal. I just never got the same level of passion for it as I did computationally. I enjoyed understanding what should happen [laugh], understanding what was the theoretical underpinning, as opposed to dealing with practicalities like, oh, you didn't store it right or, oh, you dropped in some dust. Sorting out the practical side of it didn't appeal to me for a long time.

ZIERLER: What were some of the foundational theories in theoretical chemistry that you might've encountered even as an undergraduate?

MILLER: The main one is Schrödinger's equation. As an undergraduate, I had the pleasure of taking physics and chemistry classes that focused on that, both at the graduate and undergraduate level. Another thing I encountered was the underpinning of density-functional theory, the Hohenberg-Kohn theorems, and those provided a very powerful connection between computation and useful predictions. I gained familiarity with that. You learn about perturbation theory. You learn variational principles. You learned a lot of those canonical mathematical tools that play a central role in scientific computing. Theoretical chemistry also relies a lot on classical mechanics, and I learned about hat. It's one thing to solve the Schrödinger equation to get the energy and, therefore, the throughput of that is the forces. But then once you got forces, they're going to knock around that atoms, and you get a reaction, and things move, and molecular dynamics happens. That's all to first-order the domain of classical mechanics. You learn a lot about that at the undergraduate level.

ZIERLER: Tom, when you started to think about next moves after undergraduate, did you think about industry at all? Was industry on top of any of these advances, or this is way too far afield at that point?

MILLER: I did, actually. I'd done internships. For each of my summers, I'd done internships at other academic labs. In my sophomore year, I did an internship in Florida, the University of Florida. Then I did in my junior year, University of Minnesota. In my senior year, I'd received a Marshall Scholarship so I knew I was going to go to the UK, and this was kind of my chance to maybe do something a little bit different. That was a really interesting summer. I had a couple options that I was faced with. One was kind of a policy option. I basically had emailed all of my representatives in Texas, my government representatives [laugh], and said, "I just won a Marshall Scholarship, and I want a great internship. What have you got?" [laugh]

ZIERLER: [laugh]

MILLER: Most of them either ignored me—Kay Bailey Hutchison had one of her staff members call me, and they arranged for a science policy internship that would've been pretty interesting, I think. That was one thing. The other thing was IBM offered me an internship. They were working on Blue Gene back then, kind of one of these supercomputer things, and offered me an internship there. Those were the two main ones that I can remember. That was an academic/Bell Labs-style industry opportunity at IBM, and then a more policy-oriented opportunity. I remember that feeling like one of the first branch points of my career. In the end, I went to IBM, and had an amazing time. I think that was a good option. But that was an interesting branch point.

ZIERLER: Tell me about winning the British Marshall Scholarship. Did that give you opportunities throughout the UK, or that was specific to UCL?

MILLER: It did. It basically allowed you to have support to go to any UK university. You still had to get accepted to that university—but I think people typically do—so it allows you to go to any university. I made the case to go to University College London because of the fact that David Clary, who I worked with there, was a very natural person to work with, given my academic trajectory. Like all of those prestigious scholarships, the Marshall Scholarship puts you in a cohort of just incredibly impressive people across many fields. You get opportunities. It opens different opportunities and doors that [laugh] I probably could've taken more advantage of. You could really network like crazy in that scenario, and build. I kept in touch with a number of people from that but not in a very proactive way. When I was there, I was really focused on science. In that sense, the whole point of these scholarships is to get you to go there and to make you think bigger. I eventually did think bigger but maybe later. [laugh]

ZIERLER: [laugh] A delayed reaction.

MILLER: Yeah, it was like a delayed reaction. The scholarship was very successful because I built very strong ties to the UK. That's another dedicated reason. My wife is British, so I have a very strong connection there. My collaborators are British. I have a very strong connection there. That side of it, maybe within the field, really did take meaningful root, I would say.

ZIERLER: Culturally, coming from Texas, what kind of adjustments did you have to make in London?

MILLER: It was great. It was just wonderful. London's a fun place to be young. [laugh] I definitely enjoyed that. There's all the usual stuff of living in a foreign city. One thing that was really interesting was the world was bigger in terms of career options than I had anticipated. Growing-up in College Station, Texas, it's a university town. All of your role models are professors. Everybody that you admire, and the adults that you are most impressed with are all professors. You get this kind of view of the world that that's the only thing that successful people do. You get a distorted sense of the occupational pie, I would just say. I think that really did establish my predisposition to becoming a professor pretty early on. But when I went to London, you see, all of a sudden, oh, my gosh, there's investment bankers. Oh, my gosh, what is a management consultant? I'd never even heard of that, these different things. I had another kind of period of exploration and freak-out where, oh, maybe I want to do this. I did an internship at a bank. Maybe I want to be a scientific consultant, so I did an internship there. My advisor, David Clary, was really nice about letting me spend the summers kind of doing these things. I wondered if I wanted to go to medical school, so I studied for the MCAT. I volunteered at a local hospital there, and served tea, and it was terrible, just like a usual volunteer. I went through a real period of the world is much bigger than I kind of originally thought in terms of career options. I went through a very diffuse exploration period, only to then conclude, I like what I was already doing. [laugh]

ZIERLER: [laugh] What was David Clary doing at that point?

MILLER: He had just moved from Cambridge to University of London, and for the first two years, he was at University of London, and those were the two years that I basically joined—maybe even one year earlier. But then at the end of my second year there, he accepted basically a deanship at Oxford University, the head of mathematical and physical sciences, so he was going to Oxford. That was supposed to be the time when my Marshall Scholarship after its two years was done but I had an NSF scholarship also that I could use subsequently. The original plan was to go back to Berkeley for my graduate school but when David Clary went to Oxford, I was like, oh, well, that sounds fun too. I convinced him to let me come and use my NSF funding, and I did my PhD at Oxford as opposed to coming back to the States for the original plan of doing my PhD in the States.

ZIERLER: When you were in London, that was a terminal program? That was never supposed to extend into a PhD?

MILLER: Yeah, it was never supposed to be a PhD. It was always going to be a two-year program.

ZIERLER: How do you think being in the UK generally, and Oxford specifically, just culturally, their approach to science, their approach to chemistry, how did that change what you were involved in?

MILLER: I think there's a few ways. I think one of the big ways is that Oxford was the first really top, top, top-tier university I'd spent a lot of time at. That was fun. Famous people came through all the time. Just the caliber was super high. The students were super smart. It was really fun to be in that environment. The English style of theoretical chemistry is incredibly thoughtful. It takes a lot of pride in rigor and elegance. [laugh] I tend to be a practical guy, so I think that that was something fun that I kind of honed my skills a little bit more in that direction on. I think that was very enjoyable. It stretched me in those ways, that style of academics. I would say that those are the leading order influences: the style of UK theoretical chemistry, as well as Oxford's caliber.

ZIERLER: More generally, what were some of the exciting developments in theoretical chemistry at that time when you were thinking about thesis topics to focus on?

MILLER: That's a great point. It comes down to this. How do you bring rigor into a practical domain? When I was saying one of the themes that was a kernel of my research program at Caltech, and how that led eventually to translational science, a very clear example of that did emerge when I was at Oxford. It was the idea of Feynman path-integrals, and its application to chemical dynamics. When I was doing my degree at University of London, I was using Feynman path-integrals but in the usual way: for statistical mechanics. You can basically do Boltzmann statistics, and you can use Feynman path-integrals as a really nice way to get quantum mechanical effects while using the machinery of classical Monte Carlo sampling. Usually with Boltzmann quantum mechanics, you would have to have all the different eigenvalues, all the different energies for the quantum states. You have to sum over those exponentiated energies, and that's how you get a partition function that dictates all the statistics. On its face, that's hard to do because of the fact that obtaining each of those eigenvalues is itself a herculean task, and then the sheer number of those that you would have to solve for is astronomically large in a high dimensional system. This is like a totally impractical problem to solve. What Feynman allows for in statistical mechanics is to say, well, if I just do a couple of very cute mathematical operations, I can rewrite the whole thing as a classical mechanical problem, and I have mature computational tools for sampling a classical mechanical problem that don't scale in a problematic way.

This was something that had been appreciated and effectively used in the area of statistical mechanics, namely static properties of complex molecular systems. But what about reaction rates? What about the way stuff moves in time? Can you use that trickery to also get a much more practical way to include quantum mechanical effects in the way things move in time? That was something that David Manolopoulos, a professor at Oxford, had just started playing with. They had a first paper come out as I was showing up with David Clary, and at tea time we were talking about it, and I had a lot of interest because I'd worked on the statistical mechanic side. Man, I loved that idea. I just loved that idea. I jumped all over that, and we did a lot of applications, and Clary let me collaborate with them, and that framed the early part of my career, I would say. That led to some of my major postdoc work, and also some of the early wins in my independent work at Caltech.

ZIERLER: Given the size of the data sets, and your appreciation for just how enormous these things were, was AI, just technologically or culturally, was that part of the equation at that point, or this is still too early on?

MILLER: No. Sure, you do kind of linear regression, and you kind of do data analysis and that sort of thing, which is effectively what much of machine learning is. It's just more powerful versions of that. But, no, I didn't think about AI or machine learning. I think it's a funny point that maybe because of my kind of [laugh] imprinting on the British model, I was really unexcited about machine learning and AI for a long time. [laugh]

ZIERLER: Interesting. [laugh]

MILLER: I was not one of the first people at that party.

ZIERLER: Was Oxford more of an old-school place that would've been slower to catch on to that kind of thing?

MILLER: I think that's not a fair thing. I wouldn't say that. Oxford, it's very innovative, and they're very creative and everything. But I do think that there is a pride and a style of science that is tethered to rigor. At least in my field-specific experience. Machine learning, I think, if you approach it from that prism of "Is it rigorous?", you can come to the conclusion, well, machine learning is an abdication of rigor. [laugh] It's like, well, I can't figure it out. I'll just infer it, and let the machine learn it. That was at least the way I had naively looked at it, all the way until 2018.

ZIERLER: Oh, wow. [laugh]

MILLER: All the way until 2018, I was not interested. I just didn't do it. Then I basically had a student that wanted to work on it. This is Sherry Cheng. She was able to frame it in a couple of useful ways, and we had a very influential group meeting where we said, OK, if we're going to do machine learning, let us put in as much of the physics as we know, and then the remaining part that we're going to just use brute-force computation on, let's use the machine learning to do that part. We're not going to abdicate our understanding at all. We're going to bake in all our understanding, and as a result we're going to arrive at an easier machine learning problem. We'll just machine learning the last bit, the part that we would otherwise have just brute-force computed anyways. That was the first click in terms of what we do at Entos, and how I became addicted to using machine learning in these different ways.

ZIERLER: Tom, last question for today. What were the principal conclusions of your thesis, or what did you see as your contributions to the field at that point?

MILLER: For my PhD?

ZIERLER: Right.

MILLER: I think, strictly speaking, my PhD focused on those statistical path-integral methods, and the use of them in low-frequency motions of molecules like big molecules torsions, and this allowed me to bring that rigor to unusually big systems. It was that theme again, kind of the rigor of bringing kind of path-integral physics, bootstrapping that to more complex systems like large biopolymers or biomolecules, but doing that by cleverly applying it in the torsional space. That was my contribution there. What doesn't appear in my thesis but which I also did at Oxford, which was in the long run much more important, was embracing Feynman path integrals for chemical dynamics. I didn't invent it. I was not on that first paper. But I did recognize its value very quickly, and I contributed to a lot of the early development of this ring-polymer molecular dynamics method and its use in simulating liquids and chemical reactions.

ZIERLER: Tom, that's a great point to pick up for next time. We'll see what happens after Oxford.

[End of recording]

ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It's Tuesday, June 7th, 2022. It's great to be back with Dr. Thomas Miller. Tom, great to be with you again. Thank you for joining me.

MILLER: Thank you, David. Good to be back.

Berkeley and Advances in Theoretical Chemistry

ZIERLER: Tom, we're going to pick up from at the point at Oxford when you defended your thesis. When it was time for you to start thinking about next opportunities, did you think about staying in the UK? Was that an attractive option at that point?

MILLER: By that point, I just got married right at the very end of my time in the UK, in 2005. My wife is English. But we basically planned to come back to the US. We considered whether or not to go to one position in New York or another position in the Bay Area, and we eventually went to the Bay Area. We loved the time in the UK. We had strong ties. But because she'd grown up in Belgium, the UK didn't have as strong a lure, I guess.

ZIERLER: What was it about Berkeley? Was it a professor you wanted to work with, the research culture? What was the draw there?

MILLER: In my field of theoretical chemistry, there were two absolutely eminent scientists: Bill Miller and David Chandler. They had agreed to take me on as a joint postdoc between the two of them, and it was a very special opportunity.

ZIERLER: What was your research at that point? What did you want to do when you got to Berkeley?

MILLER: I think I basically wanted to continue on the journey of taking the methods in general strategies that I'd been doing during my PhD, and taking them into ever more complex systems, so David Chandler, who specialized in statistical mechanics and the description of complexity in molecular systems. Bill Miller was just an absolute leader in terms of the more fundamental aspects of chemical dynamics. They really were two halves of the coin of theoretical chemistry in a way that was a lot of fun. Berkeley's a special place too.

ZIERLER: Being in the Bay Area, just to foreshadow to 2020, was start-up culture, was entrepreneurialism, was that in the air? Was that sort of part of your world at all?

MILLER: I'm sure it was in the air. It was not a part of my world. I remember distinctly being in Oxford. I remember David Chandler visiting one time and saying that Oxford should do more in terms of entrepreneurialism, which is kind of funny because he never did anything in terms of entrepreneurialism at Berkeley even though, I guess, he was surrounded by more of the Bay Area vibe. But I didn't [laugh] pick it up in Oxford or the Bay Area, I would say. [laugh] A very one-track mind.

ZIERLER: What was the main research that you did during your postdoc? What were you focused on at that point?

MILLER: I had this vision that I'd somehow take those two great professors, and we'd all work together on a grand problem. The reality is I just worked on two different problems. [laugh] With Bill Miller, I worked on dynamics. With David Chandler, I worked on totally different statistical mechanical things. [laugh] It was fun. We worked on the way water plays a role in driving protein folding, some of the processes of that. I worked on ways in which electronic transitions with relevance for solar energy could be better described using models for chemical dynamics. They were very fun problems I got to grapple with but I didn't quite bring the two into one grand problem together.

ZIERLER: Tom, what do grand problems in theoretical chemistry look like?

MILLER: Well, I would've settled for anything where we all just put our name on one paper, honestly.

ZIERLER: [laugh]

MILLER: [laugh] But I was just triangulating on whatever common ground I could come up with my two advisors. There were various grand challenges in theoretical chemistry. I think non-adiabatic dynamics is one. The description of strongly correlated electronic systems is a great challenge. Many people point to glassy dynamics, the way which very slowly evolving systems age, as another grand challenge. Those are fine challenges, and my work touched on some of those a little bit. But I've always been more excited about methods that enable—I think that I was always more driven by whether you actually make a better prediction. Can you actually make a better algorithm that'll execute on real chemical problems? That was a big draw for me as opposed to the deepest, the most thought-provoking, the most intractable problem. I liked finding problems that I think I could solve and developing methods that I think I could bring to bear on them, and to go after those.

ZIERLER: This atmosphere, this was a fundamental science kind of world you were in? You were not thinking about translation, applications, things like that?

MILLER: Absolutely. When I act like I'm such a practical guy, that's really on the spectrum of impractical people. [laugh]

ZIERLER: [laugh]

MILLER: But any casual observer would regard what I was doing as fundamental science. [laugh] But there's always resolution within resolution.

ZIERLER: To the extent that the Bay Area generally, and Berkeley in particular, is always ahead of the curve in computation, was that relevant at all for your postdoc, would you say?

MILLER: Yeah. I wouldn't say it was really infused with dot-com culture or any of kind of that aspect of computation. The Bay Area is just like Caltech. It's a storied university, Berkeley. It's got a huge number of fantastic scientists, and it produces a huge number of fantastic scientists. I was there basically to take advantage of that environment.

ZIERLER: What would you say your key contributions or findings were with both lines of research with each professor?

MILLER: With David Chandler, I managed to take a method that had only been demonstrated in really simple systems, low very low-dimensional systems. I showed that you could take that idea. You could scale it up. You could implement it on a real protein-folding-like problem. Something that only had been applied in kind of three or four dimensions, I was able to apply in over 100,000 dimensions, basically to a real at-scale problem, and find some really cool results about the way water fluctuations can play a very active role in protein-folding. The protein will partially fold. It will induce a fluctuation in the water. That fluctuation will kind of create a cavity into which the protein folds, a very interesting aspect of the way water mediates the folding of the protein as opposed to the protein driving its own folding. Teasing apart that story was something that I really developed new methodology to be able to do. That was one fun story. On the dynamics side, I worked extensively on taking methods that had been very effective for including quantum mechanical effects on one electronic surface, and I extended that. I worked to extend that to descriptions that would allow multiple electronic states so that if there was an electron transfer, or there was an absorption light, you could describe that. There are huge areas of chemical phenomena that are not captured by the original theory, and I worked on ways to extend the theory to those other areas of chemistry.

ZIERLER: Tom, your work on protein folding, did that get you thinking at all about human health?

MILLER: In a sense. When you have to do a job application, you basically have to say, OK, here's what I've done. Now, I need a really great way to kind of make that compelling. When I was applying for faculty jobs, I was trying to take the tools that I had developed, and say, OK, if this really works, what is something important that I could speak to? On the one hand, it led me to enzyme reactivity, which is something I worked on in my earliest days at Caltech, as well as the way in which proteins not only fold but ways in which they get across cell membranes, and they interact with channels that are dedicated to protein conduction. Those are two areas that I was led to as a result of that postdoctoral work, absolutely.

ZIERLER: Now, in terms of your career path at this stage, were you completely on the academic track? Did you think about industry at that point?

MILLER: No, at that stage, I was completely focused on academics. During my graduate years, I had the grand exploration of other jobs. But then for the postdoc, it was like, if I want to get a faculty job, there's no two ways about it. You got to focus. I was very single-minded on that. I actually applied for jobs after just one year at Berkeley. The Caltech job came up, and there hadn't been a position at Caltech in theoretical chemistry for like 40 years, so I applied for it. I was lucky to get it. I then deferred for a second year, basically, to spend that second year at Berkeley instead of coming immediately.

ZIERLER: Do you know what the backstory is, a 40-year drought in theoretical chemistry at Caltech?

MILLER: It wasn't a drought. It's just they had a group of really strong theoretical chemists that they kind of hired at the start of their career, and they all just never left.

ZIERLER: I see.

MILLER: [laugh]

ZIERLER: I thought they didn't have one. OK. That makes much more sense. [laugh]

MILLER: [laugh] I think Bill Goddard was my next-most junior colleague. Caltech had a very strong commitment to theoretical chemistry but there had been very little turnover.

A Culture of Excellence and Support at Caltech

ZIERLER: In what ways did Caltech theoretical chemistry loom in your education up to that point? Were you aware of what was happening at Caltech? Was it a center of excellence? Were you following the work of some of your senior to-be-colleagues?

MILLER: Absolutely. Caltech was one of the small numbers of schools I'd applied to for graduate school. I had come as a prospective graduate school in, I guess, 2000, and had a chance to interview with all these professors. Then I'd gone off to the UK, and then I'd gone and done my master's and PhD, and I decided to go to Berkeley for my postdoc. I had awareness of the people at Caltech from that initial visit. But, of course, everything I'd been working on for like five years up to the end of my postdoc was focused on Feynman path-integrals. That's all Caltech. The nonadiabatic transitions, the transitions between electronic states I was referring to, one of the canonical examples of that is electron transfer. There's only one name in electron transfer: Rudy Marcus. Really, chemical dynamics, Aron Kuppermann played a huge role in that. Dynamics in water and in complex systems, Bill Goddard played a huge role in that. I really knew these people from the literature, and I knew their work, and I had a huge amount of admiration for them.

ZIERLER: I've often heard it said that at Caltech for junior faculty, in order to succeed because it's such a small place, you have to be able to work by yourself. There aren't necessarily numerous people that are working in the similar field as you. For theoretical chemistry, did you find that to be the case? How highly specialized were you at that point, would you say?

MILLER: That's interesting. I was definitely more focused at that point. You do have to focus, and make your own mark, rather than collaborate a ton. I never collaborated with any of my colleagues during my assistant professor years, that's right. It does remind me of a really funny story, well, a story I find very funny. When I was a postdoc, and I was considering taking the Caltech job, I talked to another kind of luminary in the field, Michael Klein, about my concern of going to Caltech because I was scared I was going to overlap with Bill Goddard because of the fact that I worked on some aspects of dynamics and some aspects of simulation, and I was worried I was just going to be kind of too close to Bill Goddard. His answer was like, "Bill Goddard works on everything." [laugh]

ZIERLER: [laugh]

MILLER: "You can't not overlap with Bill Goddard. [laugh] Don't worry about that." [laugh]

ZIERLER: [laugh] That's great.

MILLER: That really relaxed me. You've just got to make your own mark, and it turns out to be fine.

ZIERLER: Tom, a cultural question. Did you get the sense that Caltech hired you on the basis, "We're going to give you the support you need in order for you to succeed"? In other words, are you on a path where tenure is the goal for everybody involved?

MILLER: Yeah, that's undoubtedly true, and that almost ratcheted up the pressure of it. They wanted you to succeed. They give you everything. They give you a nurturing environment. They give you collegiality. They give you ample resources. They give you great students. You've got to execute on a successful research program, and make a difference scientifically. Not everybody has the same experience, obviously. But in my experience, they give you the football, and you got to run with it. You can't blame anybody else. In a funny sort of way, that increased the pressure because it wasn't just me. They so clearly wanted you to succeed as well that they were kind of rooting for you, and you didn't want to let down the side by not achieving success. It was really what you'd hope for, I think, as an assistant professor.

ZIERLER: In theoretical chemistry generally, and for what you were interested in at that point, what are the start-up costs? What are the resources that you need to get going as a junior faculty member?

MILLER: It's basically just people and a computer.

ZIERLER: You're cheaper than microscopes and all of that stuff?

MILLER: Yeah, there's no lab renovations. We're not totally cheap because of the fact that computers are expensive, and people are always expensive, and we still have decent-sized groups. The start-up package I had then is, of course, small by today's standards. It was pretty big by those days' standards, maybe a factor of two off what a normal experimental start-up would be.

ZIERLER: On both the hardware and the software side, what are the most important systems to have at that point?

MILLER: The software is usually not an issue. The software tends to be cheap or you write it yourself or you can get it pretty easily. The hardware side: you just want a computer cluster, and that's something that kind of changes every several years based on what's available with increasingly powerful chips, GPUs versus CPUs, or on-premises versus other models. The initial thing that we went with was to a cluster that was housed on campus, and I kind of went with that model for much of my time at Caltech, and then additional centralized clusters became available, and we used a hybrid of those. I never did a lot of cloud computing during my time at Caltech. Interestingly, at Entos, we do a ton of cloud computing but it's just different days.

ZIERLER: In terms of federal support, what are the key agencies for you?

MILLER: I worked on very fundamental aspects of making and breaking bonds, and methods to do that. You can really justify yourself to almost any funding agency. I think my first grant, thanks to a kind introduction from Rudy Marcus, was from the Office of Naval Research. That was, I think, my first grant. But then Department of Energy, NSF, Air Force, Army, eventually the NIH. But it was really kind of those kind of physical chemistry-centric things of energy, science, armed services, the Army, Navy, Air Force.

ZIERLER: Tom, in terms of building a niche for yourself on the faculty, relative to what your more senior colleagues were working on, what did you decide to focus on at that point, circa 2008–2009?

MILLER: It was less a concern about differentiating from my colleagues, I would say. I relied on computing much more explicitly than many of my colleagues.

ZIERLER: Would you say that's a generational issue?

MILLER: Yeah, although Bill Goddard, of course, does everything [laugh] so he, of course, was also doing that. Actually, I think it's also unfair to say a generational thing. Aron Kuppermann was undoubtedly a very senior colleague of mine but he was extraordinarily profound in the way he used computing and embraced computing in his own work. I guess I do want to take that back. I don't want to say it's a generation thing. I think it was the reality that I wanted to position myself at the interface of dynamics and statistical mechanics. I wanted to not only describe how complex systems, large numbers of atoms, behaved on average. I wanted to describe how they evolved in time. Aron Kuppermann had done dynamics, but he had done dynamics in the gas phase, describing very small numbers of atoms. Bill Goddard did many different complex systems but there was opportunity for me to really bring together fundamental and novel methods for describing dynamics and doing that in application areas such as a protein with solvent, a battery interface, an enzyme active site, and other complex systems.

ZIERLER: Students at Caltech, undergraduates at Caltech, was computer science already the dominant major, would you say, by the time you joined the faculty?

MILLER: No, everybody was hand-wringing about bioengineering taking over the world when I joined, and so there was a wave before CS.

ZIERLER: That's what most of the students were interested in at that point, bioengineering?

MILLER: Yeah, there was a lot of interest in that, and it was pulling away chemistry majors and chemical engineering majors. I see these as kind of generational things. [laugh]

ZIERLER: When did that happen? When did the computer science dominance really start, would you say?

MILLER: That's a fair question. I don't know. It must have been in the mid-teens, if I had to guess.

ZIERLER: In the way that Caltech prides itself also on having very thin administrative walls, your expertise in computation, were there opportunities to collaborate beyond CCE on campus?

MILLER: Yeah, absolutely. First of all, when I was an assistant professor, kind of as we touched on, I didn't do a lot of collaborating on campus, I would say. Now, when I did start to collaborate, it was with experimental chemists in the department, and then it was with chemical engineers, and then it was with biochemists as well. I began to collaborate in those different areas, kind of starting within CCE, and then building out. I think my first forays outside of CCE were then into the machine learning realm, computational, and mathematical sciences, working with people like Anima Anandkumar on our machine learning projects as those were taking off. But once I started collaborating post-tenure a lot of that was within CCE, I would say.

ZIERLER: What were the research questions that compelled you to collaborate within an experimental framework?

MILLER: I just think as a tool becomes useful, you want to demonstrate its utility in a predictive mode. It's one thing to say I have a great method to explain something that you've already observed. It's very fun to then say, well, I can actually take that, and I can make a prediction. Then you should be able to make that molecule or test that prediction, and see whether or not it's true, and have a much more direct prospective version of validating that methodology and driving science.

ZIERLER: What's the value as a two-way street in terms of for you, from a computational, a theoretical perspective, seeing that validation experimentally, and what's the value for the experimentalists having you as a resource to work on these things computationally?

MILLER: I think it's a good way to frame it. I think that when a collaboration first starts, there is that sort of kind of contractual nature. Why does this make sense for me to do it, and why does it make sense for you to do it? If we can both agree that it makes sense, then we do it. Many times, it comes out that there's certain things that you can easily probe computationally that are hard to directly observe, so you can get a better understanding under the hood that you can't easily see in an experiment. Then once you understand what's going on under the hood, that can maybe give guidance to what should be done next in a subsequent experiment. That's often the times why an experimentalist would be interested in working with a computational group. Why a computational person would be interested in that is because it's a great way to see your science translate to application, and then you can actually impact important problems because of the tools you've developed. As a collaboration matures, then you're kind of jointly thinking about a problem from two different angles together. You're suggesting experiments, and they're suggesting computations, and you're just working as a team towards something bigger. I think that good collaborations might go from that kind of contractual sort of mood to a genuine meeting of the minds or kind of working together towards a problem in partnership.

Computation and Collaboration

ZIERLER: Tom, would an experimental chemist come to you as a matter of efficiency to see if something works computationally, and then using that as a basis for figuring out if it works, quote, unquote, in the real world?

MILLER: Yeah, all the time. That's a very natural mode. You can't accept all those collaborations. Sometimes, that makes sense to do. Sometimes, that doesn't make sense to do. You don't want to just run a service operation where you just run calculations or do stuff because somebody asks you to do it. Everyone's time is precious and limited, so you have to figure out whether or not that activity is going to be on-pathway with the bigger scientific program or bigger scientific mission you're trying to achieve. Many times it is, but sometimes it isn't.

ZIERLER: What would be a good example that stands out in your memory of an experimentalist coming to you, and saying, "Work this out computationally, and then let's see how well it works in the lab"? What would be an example of something that was compelling to you to do that?

MILLER: Well, one of my favorite versions of that was with my colleague Bil Clemons where we had been trying to look at how proteins get into the cell membrane, just understanding protein comes out of the ribosome. The ribosome docks at a receptor on the membrane. It threads out a protein that gets inserted into the membrane. The protein loops through. That is what leads to a folded membrane protein, which is just fascinating and cool, and has all sorts of fundamental questions connected to it. Why does it enter the membrane this way? Why doesn't it go that way? What's the biophysics of all this? How can I change that? Does biology use those changes? Many different things like that. Bil and I were discussing this and having a conversation. He pointed out that if you don't successfully thread a protein in the membrane in the right way, it's going to get degraded. Therefore, if you could predict the probability that something correctly gets threaded into the membrane, that should correlate with the observed experimental expression levels, the actual amount of protein that a cell makes. That is a pretty direct readout. If you can actually predict high probability versus low probability of correct protein integration into the membrane: the more protein they see versus the less protein they see. This idea led to a very nice connection between our computational modeling and a very important question about how to proactively introduce mutations that will drive higher levels of membrane protein expression or production, which is critical to a lot of membrane protein science. That was a great example of scientific interaction that I really enjoyed, and we worked together on that for a long time.

ZIERLER: Tom, tell me about building up your research group, the kinds of graduate students and postdocs who were both attracted to what you were doing, and who you wanted to see in your group, the kinds of life attributes, areas of expertise that were compelling to you?

MILLER: That's a very good question. I found that Caltech is really replete with incredibly smart students. I definitely was very spoiled and enabled by the tremendous quality of the students. It basically means you can do more stuff because they can execute independently of you, and you can basically achieve more as a group, and you can do better in every single one of those directions because of their capabilities. The sort of people that I really found I liked are the resilient ones — there's lots of people that are smart but not everybody's resilient, and not everybody is optimistic in the face of problems and adversities that inevitably come up. Those are the people that I really found myself recruiting. Once they're in the group, those are the people I found myself really relying on because science is just one thing going wrong after another.

ZIERLER: Right. [laugh]

MILLER: [laugh] If you haven't learned to deal with that and grapple with that and struggle with that, then you have yet to go through that necessary learning process. People that are predisposed to handling challenges well tend to be very successful, and they are the kind of people I like working with.

ZIERLER: When it was time to come up for tenure as you were taking stock of your own accomplishments at that point, what was the case you might've made just to yourself at that point?

MILLER: My colleagues at Caltech made it pretty clear that somehow in some nebulous way, you to have taught the world something about chemistry or about science. There wasn't some amount of funding you had to get. There wasn't some number of papers or where they got published that mattered.

ZIERLER: It's discovery?

MILLER: Yeah. You had to have a story where you could really say, "People didn't understand this, and because of everything I was doing, we taught them this, and it makes sense, and it holds together, and it is an advance in our understanding, and that is exciting, and I achieved that, and I'm recognized in the field with having achieved that, and that is great." I will say that my tenure talk was the one talk that weighed on my mind more than any other, because you knew it was coming for years. You knew you were going to give that talk the day you showed up on campus. You knew it was about four or five years away. That talk weighed on my mind a lot over those years because that would be the point at which you have to make your case, not only to yourself but also to your colleagues and the audience.

ZIERLER: That's such a key observation that funding, all of the metrics, what you're saying is, that is all window dressing for the discovery. It's the discovery that matters.

MILLER: It's a mark of Caltech's quality that they look at it that way. Other schools are much more focused on those numbers, and I think it's really part of Caltech's strength as an institution that they do not. You don't even put your funding numbers in the package. They don't want to see it. They want to know what you have taught the world; whether you've discovered something new.

ZIERLER: What was that story? What did you tell during your tenure talk?

MILLER: That was the other thing. You've done a lot of papers by that point. You've got to choose one to really go big on.

ZIERLER: Is that the case? Did you see it that way that it wasn't a connection through all of the papers? That you can tell one story? That's really the way to do it, to look at one paper, and expand upon that?

MILLER: Maybe it's not so much about being one paper…it has to be one project, I would say.

ZIERLER: Project.

MILLER: You don't want to go there with just a grab bag of stories. Here's three minutes on this. Here's three minutes on that. You don't want to do that. The best-case scenario is when you have one rich, interesting story that takes a great arc, and lands at something that is both true and non-obvious and compelling and a combination of all this effort. That is the more ideal situation. I didn't talk about multiple projects, even though I had multiple projects. I talked about one project. It was based on multiple papers in a series. There were methodology papers, and then we applied those methodologies to a biochemical question. It focused on understanding the way in which enzyme motions participate in catalyzing a reaction. The ways the enzyme moves during the course of the reaction: are they somehow just random fluctuations that lead to a higher probability of reaching the reaction barrier and getting over, or has the enzyme evolved to coherently funnel energy into the specific motions that carry you over that reaction barrier? That seems like a very odd question to arrive at maybe in a vacuum but there was [laugh] a really active literature on that question, where lots of people were debating which of these things happened.

It was a great situation where if you had the right tool, you could actually cleanly say, well, here's a measure of statistical correlation. Here's a measure of dynamical correlation. Here is the enzyme reaction that we all agree on as being where this effect should manifest. Let's look at these two correlation functions, now that we have developed a simulation tool that can actually access this system, and let's see which of those two results emerge. We were able to show compellingly that there very extensive statistical correlations throughout the enzyme that explained the experimental observables. But the dynamic correlations that many people had been invoking were just utterly absent. If you go one or two angstroms, they were gone. There was no dynamical correlation effect. That paper was a nail in the coffin of that idea about the way in which enzymes were working.

ZIERLER: To clarify, this is a theoretical discovery that's borne out experimentally?

MILLER: Yeah, that's right. But it was an example of one of those things where it's hard to have the experiment to directly resolve two different hypotheses. There were two different proposed mechanisms, and the experimental data was making it difficult to resolve between those two underlying mechanisms. Additional theoretical analysis enabled us to perform that resolution.

ZIERLER: Tom, in the narrative so far, it's all fundamental science. You're on an academic track. Wherever you place the deepest seed in the origin story of Entos AI, would you say it's pre- or post- tenure?

MILLER: Post-tenure, absolutely post-tenure, yeah.

ZIERLER: Even the experimental work that gave you that validation that this stuff actually is working in the real world, that didn't put you on that track even conceptually?

MILLER: No. A lot of people were like, well, have you ever thought about having a lab? Have you ever thought about doing experiments yourself? No, that didn't appeal to me because one of the things I liked about being a theorist is I could work on many different problems. As soon as you get a lab that's dedicated to one thing, it's like really hard to kind of repurpose that.

ZIERLER: You're locked in?

MILLER: Yeah, you're locked in. That never appealed to me. I would say the deepest seed for where Entos came from, there's one conversation. I'd been collaborating with Fred Manby for a long time. We had been collaborating on very fundamental science things about ways to describe—I think I mentioned the origin of the word Entos, and these embedding methods. He was visiting Caltech on sabbatical. He spent a year on sabbatical with me. We're walking around, going on one of our walks, and talking about the methods we had, and what they're going to be good for, and what methods should we develop for what important problem. We were outside Millikan Library just in that little courtyard there by the chemistry admin building, and we hit upon this idea. There were kind of the canonical things you worked on as a theoretical chemist. [laugh] You want better accuracy. You really want to say, "How can I put more physics in there in order to get better accuracy?"

What we realized is that if we kind of had put that amount of effort and that amount of creativity that we were towards the problems that sort of the field was naturally posing, if we pointed that towards what we thought was the most practical outcome, if we didn't prioritize it in terms of what the field saw as the most important challenge, but instead we said, "What if we took all this creativity and all this understanding and all this effort, and we pointed it at what would be most practically important for industrial scientists, for example, as opposed to answering the question-of-the-day academically, would we be doing the same problems? Would we be choosing the same problems?" We concluded, no, we would not be doing the same problems. We should use all that creativity and all that insight but we should be going after basically more mass-market questions [laugh] that will touch on more people and have more of a translational opportunity. That was a very important conversation. You're post-tenure. You're liberated to go after the questions that you care about. You're not handed the questions by the field as much anymore. I thought it was a chance to go after a question that had the possibility of having more real-world impact. I think that was really the first version of that. Fred and I both really embraced that notion, and we began developing methods and then software and then everything else after that that kind of followed that line of reasoning.

Post Tenure Purview

ZIERLER: Tom, the timing of this conversation, and the way you responded to it, do you see the fact that it's post-tenure is happenstance, or did tenure give you a certain intellectual freedom to pursue things that you might feel otherwise too constrained to do?

MILLER: It absolutely wouldn't have happened pre-tenure. Pre-tenure, you're laser-focused on a five-year timeline. You're like, "I don't want to think about anything. [laugh] I want to get tenure. I'm going to answer a well-defined set of questions, and I'm going to make sure that I execute on that." Post-tenure, you're suddenly like, "Wow, I'm going to live 40 more years. [laugh] What am I going to do?" [laugh] It's time to reevaluate your research program. That, I think, was part of that breathing space. You have the security of the job. You don't have to worry about that. You don't have the timeline hanging over your head. You have the chance to say, OK, now with this different set of constraints that I'd been making decisions on before, with this different set of assumptions and constraints, would I answer the questions the same way? The answer was no. You answer the questions differently because your situation's different. I think that was why it happened right on the heels of tenure.

ZIERLER: Now, initially, did you think that this was something that you could pursue within the framework of a Caltech career?

MILLER: Yes, and I did, all the way from tenure in 2013 until going on sabbatical in 2020. It was just a very abstract idea about the logic for problem selection that manifest in lots of new projects. I started more battery projects. I started more of these biology projects with membrane protein expression. I started many more collaborations at that point, trying to figure out where the tools could be useful, and where they were useful. It wasn't clear. Just because you wanted to do something that's more relevant, there's many ways to be relevant. It took a while to figure out exactly how I could sink my teeth into that aspiration.

ZIERLER: From the earlier questions where you said you were laser-focused, the discovery never ends, what were some of the key connecting points from the pre-tenure focus to the things you started to look at after tenure as you were coming up with this business idea?

MILLER: I think that the pre-tenure arc is: here's an important problem for which there is no method to solve that. I will invent that method. I will apply that method to solve this problem, and demonstrate that, and teach the world something new because that was an important problem to solve. You kind of do that. I think that that continued to be the natural way in which I framed questions ever-after, even after tenure. All of my research programs still had that shape to them. But post-tenure, I allowed myself to do that in a way that involved collaborating more. I allowed myself to do that in a way where, on the spectrum between is this the most creative method ever versus an application of that method, I would allow it to draw that balance a little bit more on the practical side of actually addressing a real-world problem. In the way it kind of played out, there was just a shift in the balance of the factors that I used in designing and choosing projects.

ZIERLER: From batteries to biology, did you find yourself talking with a wider range of scholars at Caltech?

MILLER: Oh, yeah. I definitely took advantage and benefited from Caltech. But pre-tenure, one way in which I didn't take full advantage is through collaborating with my colleagues, and it was a wonderful process to work with so many of them in different ways during that post-tenure period.

ZIERLER: From 2013 to 2020, when did it start to dawn on you that maybe you would have a full-time life beyond Caltech?

MILLER: That didn't dawn on me till after 2020 on sabbatical. I went on sabbatical with the assumption I'd be returning to Caltech.

ZIERLER: Oh, interesting.

MILLER: This was something that was evolving within me. You can kind of look back and maybe see the seeds of it. But, no, I didn't leave on sabbatical with the expectation that I'm making a leap into an entrepreneurial career. I was like, well, I need this. This is something I'm really excited about —I think I did have a hunger to somehow create a business or somehow create a company using the tools we'd developed. But I saw it as an expression of the fact that the tools work and had value. It's one thing to worry about whether or not—and you try not to worry about it unduly—but whether or not the work gets recognized; the way a paper gets cited; whether or not you're publishing enough. What metrics should I be optimizing for [laugh] in my science? Do I want to do the most students? Do I want to do the most projects? Do I want to write the most papers? Do I want the most citations? Get the most awards? One very simple answer to the question of relevance is: can I solve a problem that translates into something that can stand on its own two feet? If you have a solution to a problem that could be the kernel of a company, that's a version of success of a translational idea. I think that drew me in. I know that I did, during that time, increasingly want to have one of the ideas that I worked on in the lab make that leap to a start-up.

ZIERLER: Tom, how did your evolving interests change the research group? The graduate students that came in that are on a four- or five-year timeline, while you're rapidly changing, they come in at a time when you're doing things earlier in your career that they might be connected to. How do you manage that?

MILLER: The most important thing is you try and do a good job for your students. None of these changes happens on a dime. A graduate student will work on—over a course of five years—kind of maybe two or three lengthy projects. We might divide that on different papers but, really, you kind of do two different project areas. I think that the actual content of my research program didn't change beyond recognition. The same sort of students were coming. The students that were coming, indeed, could work on similar sorts of projects but there were just more collaborations and more ways in which those methods that were developed could be applied, at the end of the day.

ZIERLER: When you went on sabbatical, and this revelation came to you, what was it?

MILLER: I think [laugh] the basic idea is that when I went on sabbatical, you got an idea. You have a code. You like the code. You think it works. Somebody's going to be willing to license it. That's great. You're like, "Goal. It's getting out in the world, and I'm proud that it's getting out in the world. That's great." But then some time during that first year, you're like, "I think this is as good as anything else out there, and I think it actually works." The fact that it actually works is really exciting. And you recognize that the probability of the enterprise succeeding at the end of the day is directly tethered to whether or not you're going to throw your full weight behind it.

New Paths in Molecular Design

ZIERLER: What works, Tom? Is it an idea? Is it an experiment? What's the match that was lit at that point?

MILLER: The basic idea that machine learning in the way that we were doing it, and computation and the way we were doing it, was providing guidance on the design of molecules in a way that was better than other people could do. That was what we convinced ourselves. The implication of that was two-fold. You can either try to bundle that up as a software package, and try to convince everyone that if they bought your software, they would have that benefit of a better design of the molecules, or you could say, "Well, this actually works. Let's just make better molecules. Let's just do it ourselves," and take that in house, and enjoy that differentiation. It was that latter strategy that, in a commercial sense, has much higher value, because it leads to a physical product like a drug molecule; and in a process-of-doing-it sense is much more fun. Instead of running around, trying to convince people that this software is so great and they should buy it, you can actually just use it and execute with it, and actually make better molecules. Then those molecules can stand on their own two feet.

It's like making a prediction. You don't have to explain what's going on under the hood. They can see it at the end of the day, and if they the physical outcome, you've won. I think what happened during that sabbatical period was that we had the opportunity to do something that we thought could be successful. It would emerge from the science that we had been so excited to pursue and so proud of, and could actually make a difference on problems that mattered in the world. All of that hinges on whether or not you're really going to commit to it yourself. If you just hand that to somebody else, and say, "Hope it works," that's a lower chance of it actually happening than if you say, "Well, I'm going to do this, and I'm going to try to bring people around me to come and to help me make it work." That latter approach is what we did.

ZIERLER: I imagine that's a much stronger case to a potential venture capitalist—

MILLER: Yes. [laugh]

ZIERLER: —that they see how much you believe in it. You're throwing your whole career behind it.

MILLER: And every recruit. When you do a start-up, you're recruiting lots of people, and you're recruiting people that could go to lots of other jobs—senior people and junior people—and they can see that's a big difference. They can see that. It's just a great example of how risky things fail. But you can actually control the odds substantially by some of the choices that you make. By committing, and assuming more of that risk, you're making the outcome more likely of working out. It's an interesting aspect of it.

ZIERLER: So I understand the timing, when you come back from sabbatical, is the point between then and from a few months ago basically the lag time of winding down your group? Is that the objective at that point?

MILLER: I never did come back from sabbatical. My sabbatical extended from kind of April 2020 to April 2022. I initially went off for nine months, and then I bumped it out for a year and a half, and then I bumped it out for two years. But sometime in that second year, I communicated to the university that I wouldn't be returning to my full-time position. They were very supportive. We talked about that. I never did come back.

ZIERLER: Tom, last few questions, just to wrap up this excellent series of discussions. Was it a lightbulb moment when you said, "I'm not coming back," or that was also sort of a process? Was there one thing that convinced you, given all of the risk, given all of the uncertainty? Was it a process or was there one thing where you said, "That's it, I'm jumping in with both feet"?

MILLER: The funny thing about a sabbatical, maybe even a sabbatical in a pandemic, is you can allow yourself not to look at that reality. [laugh] It's like Schrödinger and his cat. You can put it in a box [laugh], you'll eventually have to look at it, and something's happening in there. You can kind of say, "Well, I don't have to worry about that question for like two years so I'm not going to worry about that. I'm just going to allow things to evolve." Then by the time you are driven by the requirements of the sabbatical limits, and by just the requirements to that question for yourself, the reality has shifted to the point where, yeah, I think I do want to do this. I think I am at that point. I don't know the exact moment where I toggled from one status to the other. But I do know that when I did take stock of the situation, and did have to reach that decision, it was no longer a close call. It was very clearly the right thing for me to do to pursue this, and I was very excited about that.

ZIERLER: Between student internships and supporting junior scholars, how do you compare the pleasures of mentorship in business with academia?

MILLER: The shoe's on the other foot, I would say. As a professor, I was tenured. I was experienced. I'd seen new students many times—the first year of students will come in, and then you will teach them, and you will help them, and you'll mentor them. In a way, I was always in that mode, providing mentorship. You do receive mentorship from your senior colleagues. But after 15 years, you've more or less [laugh] got the training wheels off. When I went into a start-up culture, I didn't know anything. I would learn so much, month-on-month, in terms of how this all worked; what I should be worrying about; what it takes to build a company. What are all these jobs, and what do they do, and how do they fit together? Just the sheer amount of mentorship and learning that I needed was very different. Like I said, the shoe was totally on the other foot in terms of who was learning. Even the people that I would hire and supposedly be a mentor for, they had worked in the industry so they're telling me how it worked. They were mentoring me. You try to be a supportive leader. You try to be a good member of the team, and a good leader of the company. But I definitely feel like I've been mentored more than I have mentored since actually entering the start-up world.

ZIERLER: Tom, what stays closest to you on a daily basis from all of your years in fundamental science, either the theory, modes of collaboration? What's the greatest influence to you on a day-to-day basis?

MILLER: I think there's two things. I think there is an underlying faith in the fact that things might seem very complicated but you can figure them out. Nothing is too hard to understand or figure out. That gives you a lot of confidence into kind of going into new fields. Sure, you won't know anything when you first get there. But building on the base of what you've done in academics and at Caltech, I always have a certain level of confidence that, once I read the books or once I talk to various people or once I dig into it, I will be able to understand it. That is a kind of life preserver that you can rely on that as you go into areas that are new and scary. I think the other thing is that Caltech imbues you with the idea that you should go after what you think are the big problems. You should go after what you see as the ambitious challenges. As you do that, you should do that with an integrity and a rigor that is rooted in good science. I think those are things that I do carry every single day, and they do utilize in the decisions and in the way about which we pursue the objectives in the company.

ZIERLER: Tom, in the way that you articulated that initial idea or realization that you came upon something where you realized you and your colleagues could do this better than what was out there generally, relative to competition, when—if not now—what's your timeline on the market coming to that realization itself? Are you already there?

MILLER: In some sense, getting investor funding relied on at least some people thinking you might be right in that. Then building a company around that involves convincing a bunch of people that decided to take that job and to do it with you that you might be right in that. Then executing on that, and getting the compounds that are exciting, is further proof to even more people that, hey, maybe that really does work. I think that you're casting a wider and wider net of the people that you might convince of the idea.

ZIERLER: Tom, one last question to tie together past, present, and future. In the way that you had that dichotomy of tenure where this was not something that you could've thought about pre-tenure, is there some threshold, you're looking out X number of years to the future, where you say, once you cross that Rubicon, you will open yourself up to new ideas that might not be available or even feasible to you now?

MILLER: That's a great question. I've become much more comfortable with the idea that I don't really know what I'm going to be doing in a couple of years than I [laugh] ever was before. That's the worst possible notion for a pre-tenure professor. [laugh] It's like the last thing you want to embrace as an idea.

ZIERLER: [laugh]

MILLER: I think that this is something that I feel very fortunate to be doing. It's something that I want to make successful and that I want to continue doing for a long time. But unlike an academic career where you have control over very long timelines, you don't have that here. By making this switch once, I have confidence that if circumstances require it, I'll be able to make a similar transition in the future. That's a very new mode of operation for me in my career but it's one that I'm much more comfortable with, and it keeps you kind of feeling young.

ZIERLER: Tom, I want to thank you so much for these conversations. There's a lot of pride at Caltech for what you're doing right now, and good luck!

MILLER: Thank you so much, David.

[END]