Can we genetically encode nearly all chemistry? | Nobel Week Dialogue 2025 | Health For All

0:00 Thanks to the excellent Swedish train system,

0:03 I was able to just walk on stage here and um tell you about chemistry.

0:08 Yay chemistry, right?

0:10 And I have three brilliant students who apparently I

0:13 just met them cover the whole gamut of chemistry.

0:17 But I want to introduce an idea that I'm having a lot

0:20 of fun with these days and it is provocative because I'd like to replace

0:24 all chemists with microbes because they truly are the best chemists

0:29 on the planet thanks to these remarkable

0:32 molecular machines that they have called enzymes.

0:35 And I've been playing with enzymes for a really long time trying to get them

0:40 to do new things that serve human

0:43 beings rather than the organisms that make them.

0:46 Now, enzymes, of course, like all proteins are encoded in DNA.

0:53 And they're the products of this incredible process called evolution.

0:59 Enzymes were made by evolution and it's really

1:03 powerful process to make yet more new ones.

1:08 There's no reason that evolution would have stopped.

1:11 And you know, so I I'm trained as an engineer.

1:14 Actually, I'm not a chemist and the chemists love to tell me that.

1:19 [snorts]

1:18 But as an engineer, we looked at the most powerful design process that was

1:24 ever developed that works from molecules all

1:26 the way to ecosystems and that's called evolution.

1:30 Now, um because proteins are the products

1:35 of this simple algorithmic design process.

1:38 You turn the crank of mutation and natural selection.

1:42 It means that they can be evolved further through this process.

1:46 And that's really unlike any other human engineered system.

1:51 I mean, think about it.

1:52 Materials, uh elevators, buildings, automobiles,

1:57 these are not the products of mutation and natural selection.

2:01 Although sometimes it may seem to be that.

2:04 And so that means that there are very

2:06 special properties in the genetic encoding of new things.

2:11 Now, uh of course, I won the Nobel Prize for figuring out

2:14 a long time ago that you can evolve and artificially select new

2:19 enzymes that do very strange new things pretty much the same way

2:23 that human beings have been evolving hairless cats and corn and cattle.

2:29 We've been making things that are not

2:31 really biologically fit but that serve our purposes.

2:34 You can do the same thing with enzymes.

2:37 But now what's happened?

2:38 So can those enzymes are amazing, right?

2:41 And we can use this relatively laborious process to make new ones.

2:46 I want to replace that and I want to use the new tools of machine

2:51 learning and uh artificial and generative artificial

2:54 intelligence because if you think about evolution,

2:57 it's very much like active learning in a machine learning algorithm.

3:01 And we take lots of data during directed

3:05 evolution because we have to screen for new things.

3:07 We throw those data into building models that then

3:11 tell us which next generation to go for improvement.

3:16 And then the great thing that happened in the last

3:19 5 years is generative AI AI which can take all

3:23 this marvelous data that are in the databases of protein

3:27 sequences and use that to help us build completely new ones.

3:33 It's still hard to do that but if you connect

3:36 uh if you connect this ability to generate new enzymes,

3:41 you know, crazy enzymes that do chemistry you can't find

3:44 in the biological world but we can start to generate those.

3:47 You connect that to optimization by evolution,

3:51 you should be able to with now robotic tools

3:55 in the next few years just push a button.

3:58 Push a button and genetically encode virtually any chemistry.

4:04 Bringing me closer to the idea of replacing chemists with microbes.

4:09 So I I want to know what you think of that proposal

4:14 and whether that would help you in any of the research you're doing, right?

4:17 Because you're doing chemistry.

4:19 Could I replace you with microbes?

4:24 [laughter] Yes, okay.

4:29 But when you say um that you can technically encode all of chemistry,

4:36 do you really mean that you can actually

4:38 do anything or are there any limitations to that?

4:42 [laughter]

4:42 Boy, I get the hard ones here.

4:44 Well, of course, we um we exaggerate to make the statement.

4:47 That's why I said nearly all of chemistry because what

4:51 are the limitations of using biological biology to make things?

4:55 You're working in water primarily, a medium of water.

4:59 It's biologically relevant but may not be relevant for making,

5:03 you know, fuels, for example.

5:05 Might be hard to make some things in water.

5:08 Uh in inorganic chemistry,

5:10 you're you're I know you're working with some surfaces

5:13 that we would not call easily accessible to enzymes.

5:17 So those are some of the detailed limitations I would say uh to this idea.

5:24 Maybe I should just say virtually any organic chemistry but that's

5:27 not true either because we made the first carbon-silicon bonds.

5:31 And you know, we're starting to add new elements

5:34 to the catalytic uh to the to the periodic table of biology.

5:42 Could I do Do you use enzymes in your work?

5:46 Uh no, but I I guess there could definitely be a possibility of doing that.

5:52 Uh my main work focuses on developing

5:54 new materials that have antibiotic or antimicrobial properties.

5:59 So fighting bio um bacterial infections without actually having

6:04 to use antibiotics or having to use as much antibiotic.

6:08 And uh Yeah, so it would definitely be a possibility of using

6:13 enzymes in that even though we haven't really considered it today.

6:17 Well, if in fact, if you're trying to replace

6:20 chemical antibiotics with these new materials that you're making,

6:24 you're you're working towards reducing resistance

6:27 that comes out of evolution of enzymes, right?

6:30 Because nature's nature solves the problem

6:34 that you pose with your new antibiotic.

6:36 She just finds a mutation in the enzymes that choose

6:39 up the antibiotic and you're back to ground zero.

6:42 So maybe those surfaces then would be

6:45 less attractive to evolution to find solutions.

6:50 So you're fighting enzymes.

6:54 [laughter] Okay.

6:53 You can say it that way.

6:57 Do you see any any potential risks with do you sing a lot of AI in research?

7:05 Like risk losing potentially this trial and error process

7:10 that has been important for a lot of discoveries.

7:15 Oh, I love that question.

7:18 One of the beauties of just doing things randomly.

7:22 I mean, people told me that's not science

7:24 but one of the beauties of doing things randomly,

7:27 making random changes and then seeing how the system which changes

7:34 are beneficial is that you actually learn something you didn't already know.

7:39 If you sit down and design, you're just using what you already know.

7:42 So yes, I see that um losing that randomness

7:48 is you could send you in the wrong direction.

7:51 But the the benefit, of course,

7:55 of losing the randomness is that you can save a lot of time because

7:59 there's so many ways that evolution could

8:02 go and most of those are totally useless.

8:05 If using data sends you in the right direction,

8:09 you might miss out on some great opportunities but you might find

8:12 some pretty good opportunities and save a lot of time doing that.

8:19 But I love random.

8:22 [laughter] [sighs and gasps] Do you see any way to use generative

8:26 AI to speed up this evolution of the enzymes?

8:29 Because the limiting factor will be the actual generation of the enzymes, right?

8:33 The mutations themselves.

8:35 So how do you see incorporating AI to speed up this process?

8:40 What's happening is it it it's really incredibly exciting

8:44 is that we have all these tools for taking You

8:49 can type out any sequence of DNA you want

8:51 and you get the physical DNA back in the mail.

8:54 We couldn't even dream of that when I was a student.

8:58 But now, as long as you can compose it with AI or evolution or however it is,

9:05 you can actually test those things.

9:08 Um and and so you're right.

9:10 The the rate-limiting step is where do you get started?

9:15 That's the hard thing.

9:16 If you want to genetically encode chemistry that biology has never seen before,

9:22 we have to know something about

9:24 that in order to get started using generative AI.

9:27 But that's last year's Nobel Prize to David Baker was for technology that can

9:32 really make some bad enzymes but good

9:36 enough where where evolution could take over.

9:41 And that's really all you need is a starting point.

9:43 And the other power, Jasper, is that there's so much data available now.

9:49 50, 60 years of sequencing DNA

9:52 from the biological world all deposited in databases.

9:56 We're learning how to use that.

9:59 And of course, high throughput experimentation.

10:02 Do any of you want to incorporate machine learning in your experiments?

10:08 Do you see opportunities for that?

10:10 If not enzymes, how about the these new technologies?

10:16 Yeah, I think for me working in analytical analytical chemistry,

10:23 we do see a trend where AI is becoming more

10:27 and more useful and something that I'm considering like also getting into.

10:34 But that leads me on to a question

10:37 that I have cuz directed evolution of enzymes,

10:42 it spans many different like fields.

10:47 And sometimes like a success to creativity seems

10:51 to be to cultivate an interest in many areas.

10:56 So, but like do you see any risks with this being like

11:05 this expert's dilemma kind of being in many different uh fields not being

11:12 like fully the the expert deep expert but maybe being knowledgeable in a lot

11:19 of things and like can you connect that to like AI also?

11:27 Right, I should be replaced by AI.

11:30 I'm not expert in anything but know how to pull information from lots of places.

11:35 AI's pretty good of that at that and maybe I'll be be replaced.

11:42 [laughter] If not by enzymes, by AI.

11:44 Well, of course we need both.

11:46 We need experts who have deep foundational knowledge in a in a particular area.

11:52 But I've seen that the biggest opportunities for science lie at the interfaces.

11:58 And and of course, that's my experience being

12:01 an engineer and seeing a problem in chemistry

12:04 and having a lab next to molecular biologists in the late in the early 1980s,

12:10 I could put together this optimization I knew

12:13 from engineering with molecular biology to solve chemical problems.

12:18 And of course, everybody thought that that was just

12:21 nuts to do that because it wasn't you know,

12:25 it was so outside of what people were doing deeply in each field.

12:32 But and you know, after 30 years of applications,

12:35 it actually has proven to be very useful.

12:39 But I couldn't have done that, right?

12:41 Without collaborating with the domain experts.

12:46 So, it really depends on who you want to be.

12:48 Do you want to be that person that bridges sciences?

12:52 Or do you want to be that person who really develops

12:55 the deep knowledge and the tools you need for example to diagnose cancer.

13:02 You need a lot of novel tools to do that for the work that you do.

13:07 Where do you want to be?

13:10 [laughter] Yeah, that's that's like a constant struggle.

13:13 There are so many interesting things that you can get into.

13:17 But you also don't want to lose like

13:20 credibility in the separate fields because they're too broad.

13:26 So like navigating that.

13:29 Credibility is something I never worried about.

13:34 [laughter] In the sense that I didn't care if people said, "Oh, well,

13:37 she doesn't know any chemistry." It just went in this ear

13:39 and out that ear and served me well not to know too much.

13:43 I think you can actually know too much

13:46 that you talk yourself out of doing crazy things.

13:50 Sometimes ignorance is bliss.

13:52 Now, ignorance is not always a good excuse.

13:56 But sometimes it works pretty well because you

13:59 can take a very different approach to solving

14:01 a problem if you know all the if you don't know all the ways it can't be done.

14:07 Yeah.

14:08 [clears throat] Okay, we talked a lot about AI and the possibilities with AI.

14:14 But there is also a lot of talk about the dangers with AI.

14:18 Do you have anything to say about that and in related

14:20 to your field since you're using a lot of AI?

14:23 So, that's a very good question um and it I could go into it for some length,

14:29 but I'll just point out yes, we evolve DNA to do new things.

14:34 And a lot of people find that nerve-racking and potentially dangerous.

14:40 But I'll just point out that human beings have been evolving

14:43 DNA for as long as we've been on the planet, right?

14:46 And every time you wash your hands with antibiotic soap,

14:49 you're leading to evolution of antibiotic resistance.

14:53 And when we put pesticides on the planet, we're leading to you know,

14:57 evolution of pesticide resistance and it and every

15:01 time we choose to breed hairless cats, we're evolving things.

15:07 So, we've had at our hands a lot of powerful tools.

15:10 And the real question is you know, do the benefits that you get

15:14 from this really clean chemistry that biology can do,

15:17 do those benefits greatly outweigh the potential risks?

15:26 Do you see the field of research in general evolving

15:29 into a state where a certain type of research conducted completely by AI,

15:33 whereas more optimization is done by humans?

15:37 So, that's a very good ending question, right?

15:41 [laughter and snorts]

15:41 Will we develop ourselves to where we don't need scientists anymore?

15:47 [snorts] Um I know some people who think that that's largely the case.

15:51 I don't believe it for a minute because we'll

15:53 just go on to ask even more interesting questions.

15:57 And I think we'll just continue to push the frontiers

16:00 and I haven't seen any asymptoting towards not needing scientists.

16:06 [laughter] I thought is time up?

16:14 Yeah, I think so.

16:16 [laughter] Okay.

16:18 [applause] [applause]