Why Building New Proteins from Scratch Is Our New Superpower | David Baker | TED

Whitney Pennington Rodgers: Hi, David. David Baker: Great to be here. WPR: Thank you for being with us. And first, congratulations on the Nobel Prize. All of us at TED were really elated to see your work celebrated and recognized in that major way. DB: And, in fact, much of the work that I talked about in my Nobel Prize address was supported by the TED Audacious Project. So I'm very grateful to TED as well and excited to be back here. WPR: It's definitely meaningful that TED could be part of your journey in that really important way and part of this important work, and we definitely want to get into all of that. But before we talk about the Nobel Prize and where things have gone since 2019 when you joined us as a speaker and as an Audacious grantee, I think it would be helpful to set the stage with some background on your work, detailing the science that was recognized here, connected to proteins. You've referred to proteins as the workhorses of all living things. You said in your talk, in fact, that almost everything that happens in biology happens because of proteins. They do everything from transporting nutrients to repairing damaged tissue to supporting our immune system. And your work, of course, has been focused around creating new proteins. Why has this been such a challenging thing for humanity to accomplish? DB: Well, for many, many years, scientists studied the proteins that exist in nature. And I think they seemed almost like these sort of magical elven runes passed down from billions of years of evolution. They're really special. They're very different from anything that occurs outside of biology because they have these very precise properties, they have these exquisite functions. And so the notion that you could design new ones that would do new things was really quite foreign. And when people tried to do it, it was very difficult. So it was both that they seemed almost unattainable and that the attempts that were made were not successful, and that even thinking about the methods for even how you would go about designing a protein with a new function didn't really exist. So I think there were a number of reasons why it didn't happen, you know, a much longer time ago. WPR: And the groundbreaking thing that you've done, the Nobel Prize-winning thing that you’ve done, is using a computer program to design proteins from scratch. Can you talk about how exactly you did this? DB: Yes. And in fact, there's been a very big transition between 2019, when I gave the TED talk, and today. So we began quite a few years ago to try to understand how the amino acid sequences of proteins determine their three-dimensional structures. And just as a little bit of background, so everyone's on the same page, the genes in our genomes each encode a protein. That's what they do. And the way they encode that protein is by specifying the sequence of amino acids in the protein. And once that sequence is known, then that specifies what the three-dimensional structure is. And so when I first began at the University of Washington, we studied how proteins actually fold up from their amino acid sequences to their three-dimensional structures. And we studied that experimentally. And as we began to learn more and more, we developed computer programs to mimic that process and to try to be able to go from pick a sequence and predict the structure. And then after we had been doing that for some time, we realized that we could go backwards, not go, like in biology, from the sequence to the structure and the function of the protein, but instead start with a new structure and a new function that don't exist, and work backwards to an amino acid sequence that would encode that new protein. The difference is that in the biology case, the proteins are encoded in the genes in our genomes and the genes of all living things. In the design case, it's a completely new protein. So it doesn't exist. There's no gene that exists. So we have to make a synthetic gene, a synthetic piece of DNA that encodes this new protein. Once we have that synthetic piece of DNA, we can put it into a bacterium and it will produce the protein. And we can see whether it actually does what we designed it to do. So the first class of models we developed were traditional physical models, where we tried to describe all the interactions between all the atoms in the protein

and how those interactions guide the protein to fold up. And we made quite a bit of progress, some of which I briefly described in my 2019 talk. Now since then, we've completely switched over to developing AI-based methods for protein design. And in these methods, we take the many, many examples of protein structures -- proteins, whose structures have been determined by scientists really over the last 50 years. And there are about 250,000 of these structures now. And so we can learn, by training AI models on these structures, we can develop methods that actually will generate new proteins with new structures. And we can condition this process on a specification of the function we want to create. So for example -- and our methods are very similar to image generation methods. So whereas you might say to Dall-E or an image generation program: “Generate an image of a giraffe walking on a horse,” or something absurd like that, and you would get an image, something like that, that represents that. In the same way, we can specify to RFdiffusion, the protein design program we have created, we can say, “Design a protein which binds to this virus and blocks it, or binds to this cancer cell and stops it from dividing." And the program will generate a new protein. And then we make it in the lab and see whether it actually does what we designed it to do. WPR: It's incredible. And it seems like there are seemingly endless applications for this. And in 2019, you shared a few of them. And I'd love to talk about those that you shared in that talk. But also, I'm really curious to know what excites you today when you think about applications for this. DB: Yes. So many of the grand challenges that I described in my 2019 talk, we've really made huge progress on and now gone beyond. But I'll highlight one example. I spoke about vaccines, and during the pandemic, my colleague Neil King here at the Institute for Protein Design actually developed a vaccine for COVID, which is approved for use in humans. It’s the first de novo designed medicine. And he's well on his way to making advanced vaccines for many different viruses, including influenza. Since 2019, as the methods we've developed have become more powerful, we've expanded the range of applications quite a bit more broadly. For example, sustainability is a new emphasis. So we're working on new proteins to break down plastic and other pollutants and polymers. We're working on new ways to fix carbon and remove methane from the atmosphere. And we're working on green chemistry approaches to enable the synthesis, creation of molecules without using toxic solvents and in much lower energy ways. So the grand challenges seemed very ambitious at the time, but as the methods have progressed and with all the brilliant people who have come here to work on solving them, we've actually been able to go well beyond now and tackle new problems as well. WPR: It's incredible. And it seems exciting to think about all the ways that this could improve life for all of us in different ways. And you mentioned the sustainability work, which was something that you touched on very slightly in your 2019 talk. And it sounds like that's really ramped up since then. How did you begin to think about sustainability work as a potential use case for this? DB: One of the advances that we've made since 2019 is in developing methods for designing proteins, which can make or break chemical bonds. This is something that happens in nature. There are many proteins which do this, which catalyze chemical reactions. And now that we've mastered that, that we can design proteins that will break bonds, problems like plastic degradation or breaking down other toxic compounds in the environment start to become things that can be approached. And it's particularly interesting with compounds that weren't present during evolution, like forever chemicals, PFAS compounds. There was never any evolutionary pressure for nature to evolve proteins to break down such compounds. So there are many problems we face today, and this is sort of a general theme. In places where nature was already trying to optimize heavily

to solve a certain problem, there's not really a need for us to design new proteins. But in areas where, either because we live longer, so things like neurodegeneration are more of a problem, or because we're putting new things in the environment, like plastic or PFAS compounds, those are the places where there's a huge opportunity for protein design, because we already know from nature that proteins can solve almost any problem. And so for protein design, it's the problems that nature didn't have to deal with because, you know, people didn't live as long or because they hadn't polluted or heated up the planet, for example. WPR: Well, it seems like that work is obviously very necessary. And I'm interested to also hear what other sorts of uses have revealed themselves in the years since you joined us. It seems like sustainability, of course, health medicine. What other areas have been surprising places that you've been able to find protein design can be helpful? DB: Well, one area that's related to sustainability, that's also become more pressing since 2019 is trying to make crop plants more thermotolerant because, you know, temperatures are rising and it's very important that major crops like rice be able to grow and thrive at higher temperatures. And so one of the things that we've become very good at with protein design is to make proteins more stable. So we're excited now about applying those methods to problems like making plants thrive at higher temperatures. Other areas are in technology. We're very excited about sensing and sort of intrigued by, you know, the ability of a human or a dog to smell and distinguish between many, many different compounds. So the way that a dog does that is with proteins, receptors in its nose that can respond differentially to different compounds that are in the environment. And we're now designing synthetic proteins that can respond to many, many different molecules. So we're very excited about building things like an artificial nose. Towards that end and for more general technology applications, we're learning how to interface proteins with electronics, because then you can have a more direct coupling of a readout from a designed protein to something that we can quickly read out on a cell phone, for example. And one example of that is our -- We're designing proteins to sense compounds in the environment that would be embedded directly in silicon nitride chips. And that's again a problem that nature never had to deal with because proteins in nature were never interfacing with, you know, with electronic devices. And we've made quite a bit of progress there. WPR: How do we ensure and what role do you see yourself playing in ensuring that what you're creating is accessible and available to a wide range of people? DB: Yes. I think we start off with the methods we developed. We make those widely available. So that's a start. We're particularly interested in enabling protein design in countries where there isn't as much advanced infrastructure for large-scale drug screening, for example. Because protein design really cuts down the cost in making proteins to deal with a challenge. For example, if you are a farmer in Africa, and you're dealing with a new type of pests, say, a fungus or an insect, we want to empower scientists and researchers in the country that has the problem to actually use our methods to develop their own solutions to those problems, which may be pretty local. As things come out of here, the -- We work -- One of the things that we do is, when we license things that we have created to other entities, we always require that there be a carveout for global health applications. And that's something that the Gates Foundation has really pioneered. And so we try in every way we can to make sure that things we develop will be as broadly applicable as possible. I think, you know, it's a challenge currently that goes well beyond protein design to motivate the world

to put the resources into ways of combating things like global warming and -- and, you know, the accumulating plastic. There's been a lot of talk, but it's going to take a lot of resources, too. And we can't really control that. I'll give you an example of something that I am disappointed in. During the pandemic, there was a lot of talk about how we needed better methods for rapidly creating ways to deal with pandemic viruses. And immediately after the pandemic or during the end of the pandemic, there were even some initiatives started to work on sort of faster approaches to develop ways to protect against new pathogens if they emerge. But within six months of the pandemic ending the world kind of forgot about it. So there are some issues with the sort of -- the way that the pull from the outside world in different areas is different depending on how much profit can be made, which is a little bit too bad. WPR: Keeping people's attention focused on this is -- DB: Yeah. WPR: Seems like the challenge of your work. Well, I think this is also a really nice segue to talking about the Nobel Prize, which is a great way to put focus on this incredible work you're doing. So late last year, of course, you get word that you’ve been awarded the Nobel Prize in chemistry. Can you tell us about how you received this news and what that experience was like? DB: Yeah, it was very exciting. And -- I got the phone call. And then I think what I remember the most that day was when I came into the lab and we had this huge party here. And that was really, really special. And then in Stockholm, I think we -- I really felt that the prize was a celebration of the work of, you know, the many people, the many brilliant students and postdocs and others I've had over the years who contribute to all the things that ended up going into the Nobel Prize-winning work and that I actually talked about in my talk. And so we asked people, the former colleagues, to come. And then I think we actually set a record. There were 185 people who came, you know, former colleagues, former graduate students and postdocs, mostly former, almost all, actually, former trainees at one level or another at different times. And that was really special. I think we set a record, and this really hit home to me when I gave my talk, my Nobel speech, and then, you know, when you're up there on stage, you can't really see who's on the audience. And then when the lights came on, I saw that there was a quarter of the room, there was this whole big section, which were all former grad students and postdocs from many years back up until the current. And that was really, really special. And I think it made it clear to me, or I guess I'd always known it, but I was really impressed upon the fact that, you know, for me, really, the scientific advances are one thing, but really, training and mentoring, all these amazing people going on to do all these really amazing things, I think that's the most important thing I do. And so anyway, and then in Stockholm, we had a number of really fantastic parties with, you know, all these 185 people, including, you know, family and extended family. And it was really wonderful. WPR: That's great. Well, David, I don't think you'll be surprised to know that we're getting tons of questions from the member community. And so I want to start to bring some of those to you, to integrate into the conversation. We have one from Les B. who asks, "How do you see technologies like CRISPR and gene editing intersecting with your AI-driven protein design? Could combining these tools accelerate the development of targeted therapies or sustainable biological solutions?" DB: Yeah, it's a great question. Yesterday I had a very long conversation with the world's experts at CRISPR in Jennifer Doudna's institute at Berkeley. And we're going to join forces, just as suggested by that question. For the problem of sustainable agriculture, we can design new proteins, but then they have to get into the plants. So we can improve, say, plant thermal tolerance. We can design proteins that, in principle, could implant improved plant thermal tolerance or protect against funguses like wheat blight. But we need to get the proteins in, and CRISPR and the IGI are really experts at that. So I think this is a perfect time to partner in many areas. WPR: Locklyn F. is curious about the pros and cons of your scientific work, being either completely open source

like OpenFold, or proprietary like AlphaFold. Can both of these options accelerate progress and possibly maybe limit public benefits as well? DB: Yeah, I'm a big believer in making everything open. And I think we find that people build on what we create. And also, you know, as scientists, really one of the goals is to really have broad impact. And the more you share, the more impact there is. And I think that's also held for the people leaving my group who are starting their own group. There are now over 100 former graduate students and postdocs who've left my group to start their own labs in the US and around the world. And I've encouraged them to keep working in this area. So now we have this amazing community. We have yearly meetings where everyone's working together, and everyone meets and shares what they've done. And things just have moved really, really quickly. So I think in general having everything open is really a big advantage. Now, if you're at a company, that has to be moderated by the fact that the company needs ultimately to make money. And so there are different constraints. But at a public institution like the University of Washington, or really in any university, I think making everything open is the right way to go. And we've seen that really from the beginning. We made -- even for our earlier generation non-AI software called Rosetta, we made it widely available. And that just created this whole community around it, which has really been wonderful. WPR: And that speaks to something that we're seeing a lot from the community, which are just questions around how to ensure that the proteins you're designing, as we think about just AI and technological acceleration in general, that it remains in the best interests of humanity, global peace, kindness. How do we safeguard against misuse in a world where some people are -- there are some bad actors looking to use things for -- DB: That's a very good question. And we are -- Let me first give you an overall framing for it, and then I'll tell you about specific actions we're taking. The things that are really dangerous, the types of pathogens like viruses, like the 1918 Spanish flu or Ebola, which obviously can cause death and destruction on a huge, huge scale. Those are extremely complex -- and because they have to do many, many different things, even with the advances in protein design, it's still very challenging to make a protein that has one function, whereas a virus has to do many, many things. And so if you want to cause death and destruction on a large scale, you don't really -- the design methods don't help you. You just go to nature and you can you can, you know, remake the 1918 Spanish flu. And so what the protein design methods are really good at today is blocking, for example, viruses. Either pandemic viruses or new viruses. The idea of making a synthetic virus is still -- Well, first of all, there's not really any reason to do it. I said at the beginning that design is very powerful where there hasn't already been extensive evolutionary optimization. So viruses, there's been extensive optimization for really rapid spreading and infection. So the first point is that if you want to -- For bad actors, there already are many, many bad things all the way around ranging from things like botulinum toxin to major viruses. And I really think the primary role of protein design will be to protect against these threats and new threats. But of course, it is possible that a bad actor really has their mind set to try and create something new and dangerous. So we had a workshop at the University of Washington a year and a half ago that we convened together with the National Security Council in the White House. And we convened a panel of experts to really think about this problem. And our conclusion was, first of all, that the current generation of design methods did not pose a threat compared to the huge threats that already are present in nature. But second, that the way to control things and to make sure -- was through the synthetic gene manufacturing step, where you go from the computer to the real world. And as I said earlier, that's a key step in this process.

And that's the one where having gating and control, or at least logging, we concluded would be really, really important. And so we are urging DNA synthesis companies to keep track of everything that they're making so that in the event that there is a suspicious outbreak somewhere in the world, you can quickly track where the DNA came from, if it was synthetic rather than being of natural origin. WPR: That's great. It seems like there is some thought to how to create this responsibly as we're entering what feels like a brave new world. And you believe that there are more benefits to pursuing this than the risks that are presented. DB: Right. And that's very clear so far. I mean, huge numbers of beneficial things have been done. Even in the area of pathogenic disease -- I described the vaccines, the antivirals, and -- So the -- I think the upsides far outweigh the downside. The other thing we're doing is we are setting up sort of a committee that's reviewing new software as it's created to make sure that -- to vet -- to make sure that there aren't unforeseen consequences. I would say overall, the risks, if you think about AI in biology compared to AI generally, I think the immediate risk from AI more generally in the form of computer viruses, things that you don't have to instantiate in the real world but that work in software -- we're already seeing problems there. And then, of course, we're seeing more broader problems, with employment and AI replacing people. I think those are probably the places where AI is going to be really a problem, a negative, rather than AI in the biological realm. WPR: When you think about the future and sort of where this is headed, we have a question from Nicholas D. If you see a future where computational design of proteins could potentially displace approaches like evolution, or will it never reach those heights? DB: No, I think that already with problems like antibody design, antibody generation -- antibodies are really -- in the pharmaceutical industry, antibodies have been a mainstay. And generally the way that antibodies are developed now is either from sort of pulling antibodies out of an individual or immunizing an animal, or screening through a very, very large collection of very large random libraries for an antibody that has the right property. I think those will be displaced by design, because you can be much more intentional and design an antibody that not only binds the right place on your target to have the effect you want, but also has all the properties needed to really be developed as a drug. So I think more and more we're going to see random selection, random sort of library screening methods, which are kind of emulating evolution, where evolution is all random mutation generation and selection, we're going to see that replaced by intentional protein design. And that harks back to what I said in my TED talk where I sort of made the point that, you know, human technology outside of biology, you know -- you start from first principles. If you want to build a bridge across a river or a flying machine, you don't go looking for a log that has the right shape, but you actually construct it from first principles. And so I think now that's becoming more and more a reality throughout biotechnology. WPR: And when you think about your work in protein development, so much of what this conversation has been focused on was the things you said in 2019, when you gave your talk. What do you foresee in six years from now when we speak again? Where do you think things will be? What will you be focused on? DB: I think predicting the future of science is far harder than predicting the structure of a protein or predicting the weather tomorrow. So I think predicting how fast science will progress is a famously impossible challenge. But I can see, broadly speaking, six years from now I expect to see many more medicines approved for use in humans. I expect to see solutions to major problems across the areas that I've described. And then I also anticipate that we'll be working in the field. We'll be working on problems that I haven't even mentioned today because I haven't thought of them. So things are changing so fast that what I really hope most is, you know, we're solving problems I can't even conceive of now. WPR: That's wonderful. Well, David, I feel like this has been a wonderful conversation

and I'm just really grateful to you for joining us. We have had so many member questions. So thank you to the members as well. Thank you so much, David, for sharing all of this. DB: Thanks to everyone who's listening and contributing questions. It's been great. Thanks to TED for organizing this. And I just want to thank again all the amazing students and postdocs and others I'm working with now and in the past who've made all this work possible, and who've actually done all the work, really. [Want to support TED?] [Become a TED Member!] [Learn more at ted. com/membership]