Interview with the 2024 chemistry laureate John Jumper, recorded on 6 December 2024 during Nobel Week in Stockholm, Sweden.

How did your interest in science come about?

John Jumper: My childhood, I grew up in Arkansas just outside Little Rock. I grew up on a farm, not really a working farm, but you know, horses, dogs, cats, that sort of thing. It was about 10–12 acres and really grew up, my parents were both engineers so always very practical, very oriented in doing this to make something, to build something. I’ve at some point, and I feel like it was probably 10, 12, somewhere around that age, started to fall in love with science and especially physics. I got very, very excited about physics, would read everything I could. I was at just the right age that you could read everything on the internet. I was born in 1985, so by the mid-nineties the internet was going well. Everyone my age remembers the AOL dial-up sounds.

I really fell in love with physics. I remember my parents being a little bit worried about whether I could really get a job doing science. It wasn’t such a thing in Arkansas, but they were very supportive. I loved it. I remember just reading this and reading and really loving also this golden age of physics, this time in which we discovered quantum mechanics and special relativity and all these extraordinary things that we almost had no right to know, these beautiful facts about the universe, I just somehow loved it. I grew up, a lovely childhood in nature, doing all these things, loving mathematics, and it was a great time.

Was there a particular person that influenced you?

John Jumper: Certainly my father, who worked at this company, BEI, that made really impressive basically parts for satellites and other applications, like a part on the Hubble and being like next to this extraordinarily, someone who worked on these important things, I think was important. As I went into college, and I was a physics major, there was a professor, Med Webster, who taught the lab section for modern physics, it was one of the first physics classes I took. He was so precise and insightful. He knew so much about physics, but he also, if he asked you a question, you gave the right answer, but you weren’t sure he would know. This was to me like this kind of perfect gentleman physicist who was really embedded in this, was really doing these things. I fell in love with that, and that really threw college.

Then I think maybe the other formative experience is I went to graduate school for a PhD, originally in physics, and it didn’t quite grab me and exactly what I was working on. My advisor was a really good person, but the research didn’t grab me. I dropped out of my PhD to basically get a job because it was too late to apply to American grad schools. When I just decided to drop out, I thought I might go back, maybe do math, maybe do physics and something else. I ended up working at this company D.E. Shaw Research run by David Shaw. They were doing biology and biology on computers, and they were trying to simulate how proteins move. I remember what was absolutely formative is they were deadly serious. We’re trying to do this hard computational problem so we understand protein so we can contribute to cures for these diseases. We are studying this protein and specifically this change in this protein because we can connect it out to lung cancer and everything else.

It was this transformative thing that you can, for one thing, get together in a team and really try and solve a really hard problem, not just kind of increment it a bit, and we’re doing this because it’s the difference whether people go home from the hospital. I think just this whole connection from what I loved and I thought I was good at and like the world of mathematics and atoms and computation, and I love these things out to health and its real impact on society and I fell in love with that. It was absolutely formative. Then I went back to grad school and chemistry almost by accident. That’s another story you should ask me about, but anyway, and really was like, let’s use computers to solve biology problems. This will really matter and people will care.

How did you get interested in chemistry?

John Jumper: Twice in my career, I’ve had complete accidents. I started with a plan and the plan was to be a laws of the universe physicist. I disabused myself of that, the timescales were really, really long. A story I like to tell, a true story, is I wanted to be originally a high energy physicist, and I did some undergraduate research with the … there’s some really good high energy groups at Vanderbilt where I did my undergraduate. We got lunch one day and I said, “This experiment we’re working on, when is it going to turn on?” This was the BTA experiment. The professor sitting next to me said, “Well, you know, I might be retired by then, it’s going to be a while” and the professor sitting next to him says, “I’m a bit older. I might not be around when this turns on.” I just decided I wanted to do a little more immediate science than that. I didn’t have quite the patience. It’s important, but I didn’t have the patience for that.

That was the first kind of shift, and I go to grad school or my original PhD and then I drop out and I applied for finance jobs thinking I was going to work a year or two as an unemployed physicist, and I end up instead working in computational biology and falling in love with it. But then the third bit of absolute randomness is that my wife, Carolyn, who’s here with me today, decided she wanted to go to graduate school to get a PhD as well in genetics, and I was going to do what I had intended in physics or math, I think more physics.

We all applied to grad schools, and if you and your wife are applying to grad schools, you need to apply to the same places, because it’s like five, six years and I’m not going to live apart from her for five or six years. We applied to all the same places, one of the places was U Chicago, but I missed the application deadline. It closed at five, I thought it was going to close at midnight, so I missed the deadline to Chicago. I said, “What’s the chance? The one place I want to go is Chicago.” A few months later, my wife has a great offer from Chicago Genetics, that’s by far her best offer. I have great offers elsewhere, but nothing in Chicago. For like a month I try and convince Chicago physics to let me apply late, and they quite reasonably said “No, maybe next year”, that kind of thing. I’m telling the story to a coworker, and he says something like, “Well, do you think you could be a chemist? I know this professor in U Chicago chemistry that I overlapped with in my postdoc.” I said, “Well, you know,” – and at this point I hadn’t taken a chemistry class since high school – I said, “Sure, I can be a chemist, how hard can this be? Whatever it takes to get me out of this problem.” He must have said a very nice thing to this professor. I talked to this professor and a week later or something they say, “We’re going to open the application for you for one day.” Then they pretty immediately accept me, and I became a chemist, so I had to learn general chemistry one week ahead of the students for the class that I was TA-ing. That’s the way I became a literal card-carrying chemist, but apparently I’ve now won the chemistry Nobel maybe for biology, but it still counts.

How important is interdisciplinarity in science?

John Jumper: I think it’s important for scientists to know really well all the aspects of their work. I think it’s not enough to me to bring together an interdisciplinary group of people where you grab a computer scientist and a biologist and a physicist or something and put them in a room. As a person, I think it’s important that you learn these fields, that you go talk to the people who are good at it. I had these formative experiences, for example, I learned computer science by going to computer scientists, lunches and seminars and learning what they care about and how they think about it. I think it’s so much more valuable, and then I would go to biologists, to chemists. I think it’s really important as a person to experience and find the experts in those fields and learn how they think and then bring it back to your work.

We both work in interdisciplinary teams but also build interdisciplinary people. I remember when we were working on this project, we would run reading groups, which is quite common, this is where you would discuss some recent paper. We made a pretty intentional decision that we would do either the reading groups or about half and half the reading group on the latest in computer science or machine learning, like what were the things people were really thinking about just from a computer science lens. Or we would think about biology and read recent biology papers. What do they care about? How do they think about it? How do they measure their data and actually read less of … we built the interdisciplinary ourselves. We tried to get the best of those fields. I think that’s really important, and we started with a team that I was the most biological of the team. I went from the worst biologists and my biophysics lab to the best biologist at the time the team started. Catherine’s a better biologist than me, and she came on later, although as a computer scientist, a team that was mainly physicists, computer scientists etc, and taught them structural biology. I think it’s a fun way, I recommend it. I think biology’s such a wide subject that it’s easier in some ways to learn biology than it is to learn computing or AI, which have a longer progression to get to the edge.

How did you get into AI?

John Jumper: I came into AI from kind of the other side, I came from the simulation computing, like, we will write down really proper laws of physics and then we shall simulate them. That’s the story we will study, we’ll write down the laws of the universe, and then we shall play it forward like a clock. Or like some grand machine. But then when you actually go and do it, you find people adjusting stuff, it doesn’t match reality, let’s change these equations, let’s make them simpler so we can compute them. Let’s do all these things. Like the way that we people study Nature isn’t as clean as you imagine it coming as we tell people in school, even in college, that these are the laws and then we simulate them. It’s really a messy empirical discussion with data.

When I was at my job at D Shaw, we had these amazing computers for simulating. They built custom computers down to the computer chips themselves and a hundred times faster than any supercomputer in the world, a thousand times faster than most at this narrow problem of simulating how proteins move. Then I went to my grad school and I didn’t have that. I didn’t want to go, I was still interested in the same scientific questions, but I didn’t have the same computational instrument as I had before, it became a really active goal. Can I replace with algorithms with machine learning? What I didn’t have in hardware, I was going to try and build with algorithms and ideas. That was my direction of travel into it.

What advice would you give to a young researcher?

John Jumper: I think a couple pieces of advice I would give. One is simply that my path and the path of many other people isn’t straight. People don’t end up doing the thing they were trained to do. We’re trained to do the problems that people knew how to solve in the past, not the new ones. I think part of it is not being afraid to be a little bit myopic, to really go after the interesting thing tomorrow to get that new skill. I think the other piece of advice I would give is to meet many communities, don’t live within this narrow discipline and community. I meet a lot of grad students who go to their lab meeting and then they go to the seminar of the department they’re in and they learn a bit, but it’s ultimately narrow. They never go to the department even down the hall or go and see the breadth of science that’s being done. They miss the opportunity to import ideas and perspectives from all over. It’s easier to learn science than people imagine. I think it’s like learning a language, your second language is really hard, learning your fifth is not so bad. If you, as you dive into more disciplines, you’ll find you have this incredible ability to learn, that people don’t realise that they’ve learned more about how to learn than they’ve learned about their subject. That makes it much easier to pick up all these other things, if you go after it, it’s worth it.

How important is failure in research?

John Jumper: I think there’s a couple perspectives of failure and we fail on many scales. I like to say if nine out of ten of your ideas as a AI research scientist are failures, you’re the most productive person I know. It’s just incredibly hard. We have guesses about what to do, and then we learn from trying those guesses. The important thing about failure is to do it quickly. A lot of what I would spend my time doing with the AlphaFold team would be talking people into the smallest version of their idea, because then everyone wants to come and say, “I have a great idea”, and they’ll describe something and I’ll say, “But that would take two months. It will be two months for you to test that idea. It’s so grand. What’s the two day version of testing the same idea?” Eventually, often you could come up with, well, if that works, then we can change this line from a plus to a minus and then see what happens or add this term.

The important thing about failure is you do it really fast. If you just accept that you’re going to have a 90% failure rate – and you should just accept that in machine learning – then you need to try ten ideas a month so you can have one success a month. It really is about how fast can you fail. I would also say that it’s about also – I tell this to a lot of young scientists – it’s not about proving your idea could never work. A lot of people get obsessed at deciding whether an idea could work or never work. It doesn’t really work that way, you’re just trying to figure out what you should do tomorrow. What has got the best chance of succeeding tomorrow? Often that’s what you’re doing today, you have the investment, it’s a good idea to try it again, try a variant of the idea tomorrow, but stop focusing so much on being right or wrong as taking the good shots that will ultimately lead to success and being very efficient with this precious resource we have of our time.

You should always learn from your failures, and you should always learn as much as you can. Really good scientists that I’ve seen, and myself on a good day, learn a lot from the failures and the successes. They extract information about the problem, and it’s not really a failure. The failure is when you learn nothing about the problem, and sometimes you do, you take your shot, it doesn’t work, and you get on with life. That’s the important part, that you know more about the problem, you’ve constrained the set of solutions or non-solutions.

Why do you think it is important to give others access to your research?

John Jumper: In terms of access, two things are important. One thing that’s important is just a clear description of how the thing works. To me that is the heart of the papers, that there needs to be a description of the thing that would let someone else build it and reproduce it. That’s the soul of the scientific paper. It’s also important that scientists have access and what access really means. Our open release of AlphaFold2 as software was really important because the community built on it in all sorts of creative ways. I like to jokingly refer to it as off-label uses of AlphaFold. We had a very clear goal and intention and problem we were trying to solve. How do we predict the structure of one protein from the sequence of that protein? As soon as it was released, we saw this explosion of creativity in the field of people doing exciting stuff. Like almost immediately someone posted on Twitter: If you take two proteins and stitch their sequence together with glycine, which is a standard thing to stick them together, it will predict how proteins bind to each other, two different proteins that bind. Oh wow. Someone else had another variant of that trick. Almost immediately people started to explore and use their own creativity.

We see this a lot recently in finding new biology with this, like one paper I really love, published in Cell, they knew that there was some missing protein and how egg and sperm come together and make what’s called the fertilisation complex. How do they recognise each other biochemically. They had a list of 2,000 proteins that are on the surface of sperm known from some other study, so they just ran all of those through AlphaFold to see which ones could AlphaFold stick up against this fertilisation complex. They found one clear hit, and actually another group in Sweden also found this hit. Then this group really tested it experimentally and showed that was the right protein. If you take it out, you won’t have fertilisation events, et cetera, doing the biochemistry of confirming this. But it’s because they did 2,000 predictions and you would never solve 2,000 protein structures. It’s a year per structure or, so it’s really doing new types of science because you have access.

Part of what I think is important about open access is let’s people use it in ways that creators don’t know about, haven’t intended, and there’s so much biology, it’s a tool. I like to say, on a really good day, maybe we’re making structural biology as a whole 5% more efficient. And that’s extraordinary, right? That’s really exciting. Of course, some people have absolutely transformative applications in their area, but it’s important to have access to these tools that science as a whole gets faster.

How important is it to use science to solve the challenges we face today?

John Jumper: I think it’s incredibly important for science to contribute to the world, to the problems, to the incredible burden of disease. We have to. Why is drug discovery hard? It’s not just because people don’t spend money – they spend a tremendous amount of money on drug discovery – drug discovery is hard because we don’t understand biology. We need to understand the science of that to build the right medicines. We need to be able to ultimately … proteins or how the cell moves atoms around on the nano scale. In a lot of the chemistry we want to do, Nature does it better, does it more efficiently, so how are we going to ultimately engineer that? How are we going to solve all these problems? Because it gives us a new type of technology and that enables us to cure disease, to engineer new things. It’s really essential. I think it’s ever more apparent to people after the pandemic how weak we ultimately are in the face of disease. How challenging it still is despite all our investments and all our technology.

Why do you think you are in the field you’re in?

John Jumper: I feel like I’m in this field for two reasons. One, maybe the biggest, is I love it. I love this challenge and uncertainty and trying to solve these problems. I think trying to solve these beautiful problems of how do atoms move? How did Nature evolve exquisite systems out of randomness? I think that’s extraordinary. I think the other reason I particularly love being in biology is it’s such a clear path to the end. It’s such a clear path that if I do this work well then someone who doesn’t care about the laws of the Universe, but might care, or does care that their relatives come home from the hospital. That to me is deeply important. I’m certain now that, maybe not yet but in the next few years, there will be people who live because of the work we did and then not a direct drug from us, but the insides lead someone else to build a medicine, lead someone else to treat a disease, and that there are people walking around because I lived my life in the way I did. That’s humbling and gratifying.

What do you think are the greatest possibilities for AI?

John Jumper: I think there are many possibilities from AI. One I’m really excited about or close to my area is that we can meet the complexity of biology. Just biology is big, enormous, wide. There are 20,000 plus places in our DNA that tell you how to build a protein right, and many of these have had incredible amounts of study by scientists, many others haven’t. There is so much in a cell that I don’t see how we’re going to get there with one at a time studies. We’re going to need to measure lots of data, we’re going to need to put this into models, we need to meet that complexity. I think one of the big promises of AI is that we will understand how the cell works and it’s hard to see how we get there without it. Then we will have some idea of what happens when we do a treatment that we will basically shorten this incredible and terrible cost of drug development. Because we understand the biology better, and if we know exactly what to go after, I think it’s more direct than right now, we don’t know what to go after.

I think the other kind of big promise of AI in this field and what David’s work is justly rewarded for the Nobel, is that we’ll be able to engineer proteins that will ultimately be therapeutics that will possibly be the dominant therapeutics and that we will treat disease faster and better. I think like local to my area, I think there’s a real promise that AI will help us solve this scientific problem, hopefully others, and we don’t know for sure. One of the things I’ll say is that we benefited from an incredible data source, the protein data bank. It was collected over many, many years and there aren’t too many of those in science. So there is a question of how much further will we be able to push this? Will we have a couple of problems that are transformed by AI and many others that got no faster? Will those problems be enough to really deliver the benefits we want? We don’t know yet. I think we should have a humility about it, but there is a promise that it can do quite a lot.

I think this prize is also really exciting because it’s a acknowledgement that AI solved a problem that humans couldn’t by any other means. It’s maybe different in a kind from some of the work that’s done with language models or image generators where they’re doing incredible outputs that look like a human could have made them, and that’s the high compliment. But AlphaFold is about building something, solving a problem that we had no other way to do it – other than experiments – but there is no way that you could do it without experiments, no amount of human thought or cognition is honestly any good at all at predicting the structure of proteins. There’s no person who thinks about it real hard and then writes out the protein structure. I think it’s the promise that we’ll solve these grand scientific problems as well.

How important has your family been to your work and career?

John Jumper: Oh, my wife is so terribly important to my career: a counsellor, an advocate, pushing me to stop doing the things I shouldn’t and pay attention to the right things. I would not have been able to do this without Carolyn, she’s extraordinary. She unfortunately didn’t get her PhD because the advisors kept leaving the university after a year and a half, and it makes me sad, but she’s absolutely been behind me the whole way and extraordinary.

Watch the interview

Did you find any typos in this text? We would appreciate your assistance in identifying any errors and to let us know. Thank you for taking the time to report the errors by sending us an e-mail.