Contact Us

Episode 12: Tackling Plastic Pollution with AI-Predicted Protein Folding

In episode 12, join Ron as we delve into the fascinating world of protein engineering and bioinformatics with special guest Daryl Barth, a doctoral candidate at The University of Texas at Austin. Daryl's groundbreaking research in protein engineering is revolutionizing waste plastic management using cutting-edge AI technologies.

Discover how Daryl's journey from materials science to biology led her to tackle the global pollution crisis head-on. Learn about the essential role of proteins in living organisms and why protein folding is a crucial puzzle in scientific research.

Explore the integration of advanced AI tools like AlphaFold and Evolutionary Scale Modeling in Daryl's workflow, and uncover the transformative impact these innovations have on traditional approaches in protein engineering.

Join us as we discuss the potential impacts of Daryl's research on waste management and recycling industries, and gain insights into the future of enzyme identification and design.

-BioML Society at UT:
-Lecture from John Jumper (DeepMind PI) on AlphaFold:
-The foremost language model is Evolutionary Scale Modeling from Meta and they made an Atlas of millions of metagenomic proteins:
-Google Colab for running AlphaFold (a faster version of AlphaFold called ColabFold) is here, maintained by Martin Steinegger’s lab:

Ron Green: Welcome to Hidden Layers, where we explore the people and the tech behind artificial intelligence. I'm your host, Ron Green, and I'm very happy today to be joined by Daryl Barth. We're gonna discuss her work on protein engineering and how she's leveraging recent advances in artificial intelligence to significantly speed up the discovery of novel enzymes with the potential to break down plastics, all with an eye towards addressing our global pollution crisis. Daryl's a doctoral candidate in cell and molecular biology at the University of Texas at Austin, with a focus on proteins and protein design. She received her BS in material science and engineering from the University of California, Berkeley, where she first became infatuated with natural materials, sparking a transition into the field of biology. Alright, welcome to the show, Daryl.

Daryl Barth: Thank you, Ron. I'm really excited to be here.

Ron Green: I'm really excited about this, too, because I really don't know that much about biochemistry, let alone protein folding and enzymes. But before we jump into all that, let's talk a little bit about how did you get interested in this field? How did you get interested in protein engineering?

Daryl Barth: Definitely. It was a bit of a long road. So as you said, I studied material science for my undergrad, and I got really interested later on in natural materials because there's a material property called toughness, which is essentially like how much energy a material can absorb. And that is really hard for us to engineer, but nature is really good at it because you need to build in gradients. You have like a soft material on the inside and a hard material on the exterior, and that really allows you to be able to be tough. So things like your teeth are really tough, I think one of the toughest materials in the world is like a limpet, something that exists on the side of a rock.

Ron Green: Okay, really quickly, because I've never heard this before, is it important that the inside is soft relative to the outside for toughness?

Daryl Barth: Not necessarily. It's just the kind of interplay of that gradient. So it's like, if you have something that's really hard and you hit it, it could still be brittle. Like if you think of a ceramic, it's like really hard to the touch, but you can break it really easily. Whereas something more like a metal, it has a bit more of an interplay between the like hardness and flexibility. And so it can absorb a little bit more energy. And so essentially natural materials are way better at, you know, capitalizing on those properties. And so I got super interested in natural materials, particularly because they're made at very ambient conditions. Like we spend so much energy and effort on making materials, especially for construction. Like refining steel takes up like, I think at least 10% of the amount of energy that the US produces a year just to refine steel.

Ron Green: That's incredible. Yeah. I did not know that.

Daryl Barth: Which is crazy. Whereas you have a spider and spider silk is the, like close to the same tensile strength as steel, but it's made, you know, in a spider at 70 degrees, you know? And so I was like, oh, this is amazing. I really need to get into biology. So I spent some time trying to transition after I graduated. So I worked in like a tissue engineering lab and then I did some more genetics, kind of like cellular reprogramming type stuff. And then happened to land in a lab at UT that focused more on proteins and proteomics. And that's been a really, really cool place to be. Because proteins are the, like the sort of cellular machines that, you know, make everything happen.

Ron Green: Okay. Perfect. So that was going to be my next question is to kind of tee up the rest of this conversation around protein engineering and protein folding. Can you describe what are proteins and why are they so important to biological life?

Daryl Barth: Definitely. Yeah. So, um, proteins, I feel like when people first interact with proteins, especially if you work out or something, you think of your like, yeah, yeah. So proteins fundamentally are, um, molecular machines. So they make most processes in the cell happen, for example. So like, if you're trying to, um, transcribe DNA, there's a protein that does an enzyme specifically. Um, if you are needing to like build muscle, right? You need a protein. So they essentially like, without proteins, nothing would happen really. Yeah. So like your cells are mostly made up of protein and lipids. So like proteins, fats, um, and then we have like a little bit of DNA and RNA.

Ron Green: Okay. So it's basically one of the core biological building blocks of life. Okay. So I understand that proteins are made from a string of amino acids. Is that correct? Okay. Can you walk us through the process of like, what does it mean to take a string of amino acids and produce a protein and what is protein folding and why is it so complicated to understand?

Daryl Barth: Exactly. Yeah. So, um, let's see, there's four different levels of protein structure. So on the first level, it's literally your string of amino acids and there's 20 amino acids. And, uh, again, if you're in like the workout culture, you've probably heard of like lysine, arginine, those types of things. So those are some of the amino acids. Um, and like the first level of protein structure is what is your amino acid sequence? And then the second level is, um, each of the amino acids has a different property, um, or kind of groups of properties. So some of them have hydrophobic residues. So they, they don't like water. They tend to be more oily. Some of them have polar residues. Um, and then some are actually charged. They have a positive or a negative charge. Um, and then others like to form more of a, the backbone of the protein. And depending on these, these kind of like molecular interplays, you get, um, different folds that happen. So like first level, you have the amino acid. Second level, you have some structure starting to come from the interplays of these, these side chains of the amino acid. So you'll have, um, the, the two, you know, that people mostly talk about are alpha helices and beta sheets. So you have the protein starting to like fold up into a spiral or kind of like laying on top of each other in this beta sheet. And then in some cases, you'll also have like disordered regions too, that don't necessarily like fold up into a specific shape. Um, so that's kind of the second level. So you have your amino acid and then it's folding up into alpha helix, beta sheet. Um, and then after that you get kind of like your, they're called a ternary structure. So you'll have this, um, like alpha helix, beta sheet, and then that folds up with another one. So that becomes like one protein domain. And then to actually form a bigger protein, you might have two that come together. And then finally the fourth part of the structure is even more. So like this is made up of two domains and then this one has like three domains and they come together to form a big molecular machine.

Ron Green: Okay. Can those different parts of the structure, can they all be produced from a single amino acid sequence?

Daryl Barth: The very, like one of the domains is a single amino acid sequence that'll fold up together. And then, I mean, to answer your question, yes, it gets a little weird. There's some amino acid sequences that'll like autoclieve into two parts and fold up and come together. You're just like, what are you doing?

Ron Green: And so the complexity, the best of my understanding is the complexity of predicting how some sequence of amino acids, which we can think of whether 20 amino acids, so there are 20 amino acids and they can be ordered arbitrarily, they're going to be converted into a protein. But that protein structure, based upon whether it's hydrophobic or other properties, will react not only to its environment but to itself. And so the angles of the molecular interactions are just very complex to predict.

Daryl Barth: Very complex, yeah. We have trouble modeling small molecules, and this would be considered a macro molecule for sure. You have all of these small interactions compounding on top of each other.

Ron Green: So, even small molecules are difficult to predict, and proteins are definitely in a large molecule. Yes. Okay. So, before we kind of segue into how some really recent advances in AI are affecting the world, let's talk about how protein engineering and protein folding predictions were kind of done until recently. I know that there were like x-ray lithography and things like that. Was it lithography or crystallography? Crystallography. Yeah. Thank you.

Daryl Barth: So yeah, X -ray crystallography and then a little bit later on it was cryo-electron microscopy.

Ron Green: Okay. Can you describe that?

Daryl Barth: Yeah, so I don't know if you've seen electron microscopy images, it essentially uses an electron as the wavelength to view whatever structure you're looking at. And so cryo is freezing, so it's like cryogenic EM. And so electron microscopy used to be used, well, I guess it still is used for inert materials for the most part. So if you, I guess if you wanted to look at this piece of wood, you could, after a while the electron would burn the surface of it, so it would be a little hard to see it. But in terms of cryo EM, the ability to cryogenically freeze something enabled us to look at biological materials, because then you could preserve the shape and actually view it under the electron microscope. So x-ray crystallography, that is diffraction of x-ray beams, and you can essentially figure out the angles that, essentially if something hits part of the protein and then bounces off at a certain angle, then you can start to build up where each atom is in space.

Ron Green: You can infer the positions or the structure of the protein by the deflections of those x-rays.

Daryl Barth: Yes, yes. That's actually a crystallography, and then cryo-EM is literally like a picture. But since you're using an electron, you can get to the resolution of the wavelength of the electron. So you can actually start to see really, really small objects when you use electron microscopy.

Ron Green: So these are traditional methodologies, but my understanding is that it's fairly slow, right? I mean, somebody might spend years trying to understand how a single protein is folded. Is that correct back in the day?

Daryl Barth: But particularly earlier on, and a lot of that is actually crystallizing the protein is the hardest part. So particularly for like x-ray crystallography, you need to actually have a crystal to be able, because you don't need, if things are moving around as you're trying to get beams off of them, then there's no way to back calculate where the thing is.

Ron Green: I definitely didn't understand that. So when you're going to crystallize the protein, what does that mean?

Daryl Barth: So I've never actually done this before, but essentially it's like freezing it. So you've made the sugar crystal thing as a kid like rock candy, that kind of thing. It's a very, I think it's a pretty similar process to that. You just have to like very slowly go through lots of different steps to try and freeze the protein without denaturing it. Because a lot of proteins are quite sensitive. They're like heat sensitive, they're UV sensitive. So there's all these different things that can make it unfold. So denaturing a protein is when it is unfolded. So it's active in a particular position. Like particularly for enzymes, the shape of the enzyme, as well as its like chemical nature is important for its function. And so if you don't have that shape, then the protein won't be doing what it's supposed to be doing. Okay.

Ron Green: So the shape, so this kind of closes the loop and helps a bunch. No, this is fantastic because the shape of the protein determines its behavior, its function, how it interacts. So being able to predict what that shape would be will allow you to do things, like for example, within your work to predict enzymes that might be able to break down plastic. Okay, I would love to transition. Let's talk about some of the big breakthroughs. I'm familiar with alpha fold, but for our listeners, maybe just real briefly describe how does alpha fold, how are you using it on a daily basis to help in your work?

Daryl Barth: Totally. Yeah. So how am I using it? So a lot of it is the trying to see the structure and the function. So I guess in research before, literally like three years ago, which is kind of crazy.

Ron Green: It's crazy.

Daryl Barth: Yeah. If you actually wanted to see your protein, it would be arduous, you know? And so you would only want to actually look at the protein if you knew this was something like useful that you wanted to characterize further.

Ron Green: You mean actually take the time to try to really study that protein?

Daryl Barth: Yeah. And there were, I mean, so before AlphaFold, there also were other methodologies that we talked, I don't know if we should talk about that. No, absolutely.

Ron Green: Let's talk about, so I've got in my notes here, you're talking about the critical assessment of structured prediction contest. Okay, yes, which has been going on for several decades now.

Daryl Barth: Yes, several decades. So this whole field was really launched, I think in the late 50s, by a scientist named Christian Anfinsen. And they call it Anfinsen's dogma, which is very fun. But he was pretty convinced that all you needed was the amino acid sequence to determine what the protein structure would look like. And so he actually won the Nobel Prize with another scientist in the 70s for his work on this. And so he kind of launched the field and people from him were like, Oh, I guess maybe you just need the amino acid. And so like the amino acid sequence, and so people started really digging into this. And to kind of converge the field on it in 1994, they started CASP, which is the competition that you mentioned. And so every two years, it was like the protein structure Olympics, a bunch of scientists around the world would get together and try out their models on a, you know, a new set of proteins where like, they had been crystallized. So the people, you know, running the competition knew what the structures were, but everyone testing their models didn't know, didn't know. Yeah, I didn't know. And so they were trying to see how well they did. And the first approaches were all, you know, physical models. So like, like we talked about the amino acids earlier, all of the charged hydrophobic, etc, residues, how do you model them in such a way with the physical knowledge that we know, to try and get the right structure from all these different interactions. So that's how that started. And it had, you know, kind of like different jumps back and forth. There was some progress, and then there were plateaus and progress in plateaus. But really, it wasn't until alfafold came out around, the very first iteration was 2018, but the big deal was after the transformer was invented. And in 2020, alfafold just blew everyone out of the water.

Ron Green: I remember the news, it was really amazing because even though I really don't know much about this field, I do know that protein folding is considered like one of the grand challenges within biology. And alpha fold really, really kind of blew away, I believe, all previous models in its ability to predict folding. Is that correct?

Daryl Barth: It did, yeah.

Ron Green: And then two years later, I believe alpha fold two was released. And my understanding is that was, again, another sort of dramatic improvement in prediction. Is that right?

Daryl Barth: It was. Yeah, it was like completely dramatic. And I think the 2018 one was like some progress, but not as dramatic in 2020, you know, some people would consider protein folding to be solved or at least very much on the way to being solved, you know, like the first 80%.

Ron Green: So I remember reading that too, that some people were arguing that, okay, this, this really difficult problem that's been around for decades is solved. And then some people were a little bit more skeptical since alpha fold two. And now I guess the competition will be held again this year.

Daryl Barth: Um, I'm actually not sure

Ron Green: But my guess would be there's probably general consensus that this new approach is, if it hasn't at least solved the problem, we're well on our way. Is that a fair assessment?

Daryl Barth: Definitely what it feels like, like the paradigm shift is there.

Ron Green: All right, so I'd love to... All right, so going from using x-rays and electron microscopes and things like that to now just interacting with a software tool, can you take us through that process so you have a sequence of amino acids, which in this instance I think is literally you're just saying it's a sequence of letters, and each letter is one of 20 letters that represents that amino acid. So I don't know if it's A, B, C, D or something like that, but you take that sequence and then what?

Daryl Barth: Yeah, so it's to the point, I think they have an API, there's a few different ways that you can do it. A lot of people have set up Colab notebooks, if you've heard of those. Oh, absolutely. Yeah, yeah, so there's a Colab notebook online that you can go to, and you can copy and paste your amino acid sequence into it, and then click through the cells, and you'll have a folded protein.

Ron Green: That's just unbelievable. How long does it take? Does it matter the length of the amino acid sequence, or is it always the same amount of time?

Daryl Barth: It definitely matters the length of the amino acid sequence. When it first came out, it was really slowing down I think after about 1,500 residues or over like 1,000 to 1,500.

Ron Green: Wow, that's a lot longer than I thought you were going to say. Yeah. That's already pretty impressive.

Daryl Barth: Yeah, so a really small, or in the work that I do, a pretty small protein is about 200 amino acids, and then median size, maybe between 800 to 1 ,000, for monomers. So we're talking about just one amino acid sequence to that initial domain that we're talking about, yeah. They can get huge, though. I've seen some protein complexes people have folded where they're like 20,000 or even more, residues all put together. So there's some really incredible work that people have done out there, even just with like playing with all the parameters and folding multiple proteins together and stuff. So that's really remarkable if you wanna go just like look at pretty pictures of proteins. That's definitely possible. But you were asking,

Ron Green: Yeah, on your day-to-day. I mean, because it sounds like, I mean, it just, it's kind of, to me, it's actually one of the realizations of the promise of AI that we've been talking about for decades, right? This idea that AI systems wouldn't just be intelligent and help you with your day-to-day work, it could actually advance different fields of science. And it sounds like biology and protein folding, your field in particular, it's really making an impact. So on a day-to-day basis, within your lab, are you all exploring vastly more sequences than you would even have imagined like five years before?

Daryl Barth: Absolutely.

Ron Green: And so let's talk about that. And specifically, let's segue into enzymes and plastic decomposition and your work there because unbelievably important. Let's just talk about how you're leveraging it and how it's helping you make progress on that problem.

Daryl Barth: for sure. Yeah. So to kind of contrast, I think, well, actually, in my own personal experience, kind of as I've come into this, AlphaFold has been a tool, which is kind of wild. But even as I've been working on it, it's become more and more available and like possible to work with it, I guess. And a lot of other softwares have started piggybacking off of it. So I think the most compelling advance to me is there's a kind of an old school algorithm called BLAST. And it essentially allows you to compare amino acid structures or like amino acids to amino acids. So you have your sequence and then you can look through huge databases of all known proteins and see like what other species have the similar protein and kind of like line that up.

Ron Green: to find the most similar match.

Daryl Barth: Exactly. And so now that we're able to fold, you know, X amount of proteins at a time, you can now do that with structures, which is like kind of mind blowing and really cool. And so it allows you to kind of go, if you're trying to find similar proteins in the kind of mixed bag of whatever evolution has created in terms of proteins, you can now use sort of two different approaches. You can use a structural search approach or you can use a kind of sequence search approach, which is neat. So it's sort of like double your ability to search. That's important for me because there's a few known enzymes that have been shown to be effective at degrading plastic. And so one of the like research branches that I'm going down is using those proteins that are known and trying to find other iterations of them in areas where you'd expect proteins with the kind of qualities that we'd want. So one thing is when you're trying to break down plastic, typically, if you can raise the plastic to a higher temperature, your enzyme will like, the plastic will loosen to some amount, there's this thing called a glass transition temperature that it goes through. And that means that the molecules in the plastic have more freedom to sort of move around. And that in a way like increases the surface area and the ability for the enzyme to like get there and actually eat at the bonds.

Ron Green: Oh, that makes complete sense.

Daryl Barth: Yeah. And so, but to be able to do that, you need an enzyme that can handle that heat. And so that's like typically around 50 to 70 degrees Celsius, tends to be the sweet spot there. But a lot of enzymes, you know, operate in your body, right? Or more in ambient conditions. And so when they get to 70 degrees, they're like, they start denaturing and that doesn't work out. And so what we'll do is we'll go to data sets that have been grabbed from kind of extremophilic type places. So deep under the ocean with hydrothermal vents where there's microbes that can exist at like a hundred degrees Celsius. And we'll get the metagenomic data from that. So all the DNA sequences predict the proteins that are in there and then search through those protein mounds for enzymes that look like enzymes that we know that work on plastic, but that might be more thermostable.

Ron Green: All right, let me make sure I understand, because that just sounds incredible. So you're basically using the search technique BLAST to go look at collections of DNA sequences from these organisms at the bottom of the ocean, but that are near these thermal vents. So they have incredibly high temperatures, because we know they obviously can survive there. And so then you're using that to identify potential sequences. And I do have a question about how you go from DNA to amino acids in your second. But you're using that to identify potential candidates from amino acid sequences that would operate at high temperatures to break down plastics.  

Daryl Barth: Yeah, so that's one way of doing it. Yeah. And so in the past that that approach has been available I think for well, the metagenomic data is more of a newer phenomenon. So being able to have huge pools to look through is is newer probably like within the last 10-15 years that field has started to really pick up But in terms of being able to do that blast search that's been around for you know, like 20 plus years at this point So the really cool part of that is like oh then I can find I can find with blast potential structures that work and then I Can fold them and I can then compare those folds to the known enzymes and be like do they work, you know And so I can you know, basically narrow down from like 500 potential enzymes to maybe like 20 That might be the ones that are working Before I spend a bunch of money like actually expressing them and cloning them and seeing if they work

Ron Green: makes perfect sense. How do you go from a DNA sequence of some organism at the bottom of the ocean to the amino acid sequence that you may want to explore as an enzyme candidate?

Daryl Barth: Yeah. So that's translation is like the kind of name, which is actually hilarious now that I think about that. It never really came to me. That's hilarious. Translation. Exactly. Yeah. So there's three DNA, so three nucleotides that code to an amino acid. So that's like a language. There's tables where you can go three nucleotides to one amino acid.

Ron Green: Each codon transforms into an amino acid.

Daryl Barth: Exactly. Yeah. And so you're like searching through this genome that you have. There's like a start codon and a stop codon. You see the start codon and then you go from there and every three nucleotides becomes an amino acid.

Ron Green: So you don't have to go through DNA to RNA to amino acids. You can go directly from the DNA. That was one part that I wasn't sure if you were going through that full transcription process. So all right. So I'm really fascinated about this next part. So you have some amino acid candidate. How do you reconcile the theory, the prediction, of what, let's say, alphafold says it will do with the reality of what it does? How do you confirm that? How do you test it?

Daryl Barth: Yeah. So a lot of that is yet to lay it into carbon, as people say. Yeah, honestly, that's probably the next frontier is figuring out actual function. So you can, like how people try and do it is sequence similarity, structural similarity, how close is it to a thing that we have tested in real life is kind of the basic. So if you're, and then once you, when you're trying to get past that, that's when you actually have to go into the wet lab and make the thing that you want to see what it does and then test it and see what it does.

Ron Green: And so you actually go and make it, and then you, do you just expose it to plastics at high temperatures and then see how it reacts?

Daryl Barth: yeah there's, there's a couple of ways of doing it, but yeah, that's, that's basically what it is. Like you'll, you'll cut a whole punches of the plastic and put it in little Eppendorf tubes and keep them at different temperatures and take, you know, time points, you know, over days. And then there's this thing called an HPLC that you run it through and it's called high performance liquid chromatography. And there's a bunch of different ways of doing it, but you can basically figure out the, whether you're seeing the breakdown products of the plastic in solution. And that'll give you an idea of like how quickly your enzyme is acting or whether it's acting at all. So.

Ron Green: That is just amazing. Are there any, like where are we in the process? Are there any strong candidates right now that are showing promise?

Daryl Barth: Yeah, definitely. And this field has been going on for a decent amount of time. I think early 2010s and actually even earlier people were interested in enzymes that could break down plastic, but not as intensely as they are now. I feel like a lot of papers that you see nowadays, they're more industry type focused.

Ron Green: Is that mostly because there's increasing confidence that this can be done?

Daryl Barth: Mmhmm. I think that and then also demand for it, like the, I feel like in, you know, before the 2000s, the plastic problem, it was, it was kind of like global war mania. People were like, eh, you know, whereas nowadays it's, it's really gathering steam and there's funding for it, which is a big part of it as well. Um, yeah. So in, it depends on the type of plastic you're talking about. Um, so I think there's something crazy, like over a hundred million tons of plastic produced each year, which is like kind of unfathomable. Like I can't imagine. Yeah, it's unbelievable. Yeah. So there's that much produced. And then of, of the plastic that's being produced, there's about six types of plastic that make up the majority of it. Um, by far the most, um, is a plastic called PET, which is polyethylene terryphthalate. And that's what you see in water bottles for the most part. Um, and then after PET, you have, um, there's like polystyrene, styrofoam, um, polypropylene, which is a bit harder plastic, polyurethane, polyethylene, and then nylon. Um, or like the, the full six. Um, and in terms of PET, the bonds that kind of hold together the polymer there are a little bit easier to attack than the other types of plastic. They're, they're more similar to, you know, bonds that you would find in biology. And so there are more, um, instances of enzymes that can take down PET, which is cool. And so people really focused on it. Um, the very first one that people found that they actually termed it PETase was in, uh, I think 2015, there was one before that, which was called a cutenase that actually came out of leaf compost, which is kind of wild. Um, but the, uh, yeah. So in, in 2015, outside of a water bottle plant in Japan, um, there was a microbe that was, um, isolated called Idionella sachianensis. And they found that the microbe was able to live exclusively on PET plastic. And so it was able to like extract energy from it and actually metabolize it. Okay. Pretty incredible. That is incredible. Mm -hmm. And so, um, from this microbe, they were able to figure out which enzyme was the most responsible for the PET breakdown. And that became, um, a big PETase. So it's called IS PETase after Idionella sachianensis. Okay. Um, yeah. Uh, so that, that was kind of like the first instance of PET. And so between the leaf compost cutenase and the, the one from the outside of the water bottle plant in Japan, you have these like two pretty amazing enzymes that people have then, um, done some pretty intense engineering on, um, to like try and make them more thermostable. And so literally, I think it was last month, um, there was a new PETase released called turbopetase. Um, that actually looks like if you put it in a recycling plant, it would be able to accomplish, you know, breaking down PET into its monomers at a scale that you would, you would desire if you were actually trying to do this.

Ron Green: Oh, that's incredible. So really close to having something that might be useful at industrial scale.

Daryl Barth: Definitely, yeah. And there's actually a company in France who used the leaf compost cutenase one, and they've been working on it for I think like almost 10 years now. And they have a pilot plant that I think started up a few years ago, and they're scaling it up. So there's actually like one instance in the world, but one that can actually degrade PET. So in terms of that one plastic, PETase, things are going pretty great. There's a lot of hope. For the other plastics, the road is long. It's early days. But there's possibilities though. For each of the plastic types, there are either like known organisms that can eat it, or there's been like instances in the past where you might be able to work on it, if that makes sense.

Ron Green: That's unbelievably fascinating. Yeah, that's why. Wow, that I am just so impressed and I'm just so personally so excited that we're at this stage now where people like you who are doing important work are able to leverage AI in a way that I think a lot of people who listen to this are going to be really surprised by just how much it's impacting bleeding-edge science.

Daryl Barth: Oh, it's been amazing.

Ron Green: Okay, this conversation has been unbelievably fascinating. We'd like to wrap up with a fun little question. If you could have AI automate anything in your life, what would you go for?

Daryl Barth: Oh, anything in my life. Yeah, anything. Oh, not just with work. Oh, interesting. Yeah, it can be work. Oh, wow. Anything in life. Yeah, that's, that's pretty cool. I think personally it would be time management. It's a big thing, which I know, which I know is coming. I'm definitely like scheduling in and like figuring out what to prioritize. Yeah, that, that would be huge for me. Um, but in terms of the fields, um, yeah. And we're kind of coming up upon it. It sort of goes back to the function aspect. Um, like we spend a lot of time, like the wet lab pipeline is really intense. Um, and also expensive to figure out if like these enzymes are actually working.

Ron Green: we're actually producing the enzymes.

Daryl Barth: Exactly. And testing them.

Ron Green: Okay. Exactly.

Daryl Barth: And so if we had a way of being able to more efficiently or correctly characterize the function of a protein, that would be amazing. And there do seem to be, people are trying to get there, people are training classifiers off the end of alpha folds, so you have your structures come out and then they'll have a classifier of different similarly functioning proteins and so you can say, oh, does it fit in this pool or this one, that kind of thing, so that's sort of happening. And then there's also large language models that are just taking in the amino acid residue and there's some interesting stuff happening there, but in terms of actually being able to say, here's structure, here's residue, this is what it does.

Ron Green: Then, just to be clear, that next step, there's one thing taking an amino acid sequence and predicting how it will fold. The next step is, how will that actually behave?

Daryl Barth: Yes, how does it interact?

Ron Green: How does it interact in solving that?

Daryl Barth: Exactly, which is huge. And we're approaching that possibility on the small molecule scale. I did want to mention this. There's some really exciting stuff coming out recently. There's something that came out, like the code came out for it, I think literally yesterday or a few days ago. It's called Rosetta fold all atom. And so it's essentially like an alpha fold, but instead of just looking at proteins, it also incorporates small molecules and like ribonucleoprotein, right? So it has RNA in it. I think I said that wrong. But yeah, so essentially it's like RNA in complex with the protein is how CRISPR functions. And so you could ostensibly fold that whole complex using Rosetta fold all atom, which is like crazy. Yeah. So people are like working on it and trying to get there, but that next step, the sort of like, okay, we have the thing, how does it interact with its world is the kind of next step. I think that's the frontier we're trying to move.

Ron Green: We'll make sure to put links in the show notes so that people can go access all the different tools and technologies we talked about today. Well, thank you so much, Daryl. This was fascinating. I mean, I can't tell you how excited I am about your work and how you're leveraging AI and for taking the time to chat with us today.

Daryl Barth: Thank you. Thanks for inviting me. This has been really fun.