Yeah, I mean I think the phenomenon that you're describing I think is a a r really generic set of features of interesting research problems. Yeah, maybe when you know, maybe I should have said something that is like profoundly like not interesting.
SPEAKER_02Well that's awesome.
SPEAKER_06No, I don't I don't know, I don't know about that.
SPEAKER_02But it's just big words to like describe something very straightforward.
SPEAKER_06It's just it's just that like, you know, in I I would say in the fields that I'm working in, there's a confluence of a set of interesting opportunities that have been enabled by the combination of like the physical theories and understanding, you know, reaching a level of maturity where we can formulate questions that are interesting, coupled with advances in computation and in machine learning that are allowing us to approach these questions with a level of computational sophistication that previously was not available to us.
SPEAKER_04So Grant Roskoff. Rotsk, Ross, Rotzkoff, yeah. Rotzkoff, assistant professor of chemistry at Stanford. Yes. Nice. Nice.
SPEAKER_06Thanks.
SPEAKER_04So I was digging a little like I mean I can just ask you open-ended, or I can um or I can like jump into what interests me like from what I read from your biopage.
SPEAKER_06Yeah, please jump in.
SPEAKER_04Um non-equilibrium dynamics, right? And like very specifically cell I guess mitosis, right? Signaling, signal transduction, right? Um what is it like? And there's like it's it's sort of general on there, like there's a lot we don't know, right? There's a lot we don't know. Um But like, what does that mean?
SPEAKER_06Well, so you know, living systems are are extremely far from equilibrium. So when we think about driving systems, you know, away from like a stationary state, from like the state that they come to rest in, um Which would be death. Which would be death, right? And there's a there's a a widely used and famous quote from Schrdinger that equilibrium is death. Um so that sort of like stationary distribution is the one that we understand well. So the the whole framework of um you know of classical physics around explaining equilibrium, so statistical mechanics in particular, is really concentrated on a setting where systems are at rest. And and we understand extremely well how to predict properties of systems that are at rest. So we we know how they respond to external perturbations, we know how the molecular scale informs macroscopic material properties, and in principle, we can calculate a lot of these things with extreme detail. Um however, once energy starts coming into the system, so once the system is coupled to something that drives it away from equilibrium, we lose a lot of the perspective of you know, how do we do calculations that explain the properties of different materials? How do we do calculations that um explain what is what the system is going to dynamically evolve into? Um, one context where that is really very important is in biophysics. So, you know, cells and the organizational principles of cells are um you know highly dynamical. So cells don't look the same from one second to the next. They evolve a lot um as a function of time. And most of that evolution is driven through environmental perturbations and energy consumption. So metabolism in cells is something that that drives forward a lot of the um the organizational structure within the cell. Oh, that's interesting.
SPEAKER_04But keep going.
SPEAKER_06So so yeah, I mean the the basic questions you know that my lab asks are kind of like a level down from the biological organizational principles. Like we're we're interested in kind of the first step of this, which is how do the emergent properties of biological materials.
SPEAKER_05We're gonna get into it then. I could tell from what I could tell. This is gonna be good.
SPEAKER_06Yeah, okay. Yeah, uh so yeah, how how do the how do the material properties of biological systems emerge from you know the way in which they're assembled, you know, through uh through these non-equilibrium dynamics, how forces inform what types of structures uh arise.
SPEAKER_04Oh, this is so cool.
SPEAKER_06Yeah.
SPEAKER_04Okay. I'm so I'm sorry, can I jump in? Yeah, please. Okay. Uh so I read the thing, right? And the first thing I think of, well, isn't that what fluid mechanics is? Like that's the math of moving systems, right?
SPEAKER_06So to an extent, yes. And like um, you know, the basic tenets of fluid mechanics are kind of like the organizational principle at the macroscopic scale. Yeah. But fluid mechanics differs from the type of dynamics that we study in that it doesn't resolve molecular detail. So we're interested in how you know the individual components of a system, how the discreteness of those individual components informs the resulting structures. Whereas in a fluid, you can sort of, you know, you can discretize infinitely and you'll have something that's really just a continuum, right?
SPEAKER_04So the math of fluid mechanics is based on like a like a simple fluid that is, that is, there is a molecular component to it, but you're you're addressing the action of the whole collection of molecules.
SPEAKER_06Yeah, so yeah. So the the way like fluid mechanics is a continuum theory in the sense that it does not resolve the individual molecules that are you know that are in the fluid. It only assigns to, you know, a a density of fluid a a collection of properties that are related to those underlying molecules, but it never resolves those molecules.
SPEAKER_04Well, but it's also sort of not, it's different because it's not. So like I just talked to a physicist the other day, Epic, and we're and we're like going into strong force, weak force stuff. So we're not there. We're not a strong force, weak force, but we are somewhere around like covalent bonding, pi bonding, right?
SPEAKER_06Like these are the types of those are the types of interactions that govern the way that individual discrete molecules interact. Right. Which is actually you're a chemistry professor. Yes. So like you're very exactly.
SPEAKER_04We don't know how molecules move. We we understand some aspects of the Trevor Burrus, I guess we know a lot about how electrons move, and we know a lot about how fluids move. Aaron Powell Yeah.
SPEAKER_06I I I think it's fair to say that we know a lot about a lot of these things, but we also don't know a lot about a lot of these things. Okay. So you know, for electron dynamics, you know, it's it's actually very hard to predict in great detail the way that um you know electronic uh activations and electronic excitations evolve in physical systems. Like it's you know, it in principle, like there are a lot of things in physics right now where in principle we know the answer. Yeah, yeah. We know that um the electronic Schrodinger equation governs the way that electrons move, but on the other hand, like solving that for non-trivial systems is impossible. And it's similar with a lot of classical systems where we have large numbers of interacting degrees of freedom that are coupled in complicated ways. And in principle, if we had you know infinite computational resources, we would be able to actually solve an evolution equation that says, okay, here's here's how this is going to be.
SPEAKER_04What's the equation of a tree? Right. Right. Yeah.
SPEAKER_06Like if we had if yeah, and I mean it's some, you know, with some level of course graining, like we in principle could do that. Yeah. But but it's not informative because we can't reach those scales. We can't reach the relevant length scales. Yeah. And so you have to take more of a statistical perspective on the way that these things work. What do you mean? Um so you know, a lot of what we do is thinking about how do we access the the relevant information that governs the long time scale dynamics of systems that are driven out of equilibrium or are you know moving between different uh stationary states. That's something that we can't just directly simulate because solving the equations would require an amount of resources that you know exceeds uh any, you know, any amount of money in the world, or like a computational time that exceeds the lifetime of the universe, right? Like things like this.
SPEAKER_04So it's about which variables matter.
SPEAKER_06It's about which variables matter, and also, yeah, so you know, how do you do appropriate course graining, but also you have to kind of push the systems into states that are relevant and interesting and figure out systematic ways of uh of quantifying how far did you push so that you can um reweight everything back to like the correct probability distribution. So we take a very statistical perspective on it. And that's kind of like a common approach. It's called an important sampling, and statisticians use it very heavily, but the chemical physicists are kind of the ones who have pioneered like the biggest applications of these techniques because they're very focused on understanding the statistically rare events that are chemically important. So something like you know, the breaking of a bond in a chemical reaction. Like that's something that happens on a long time scale relative to what we can simulate. It's something you know that like you know, for uh Well I don't think it would be the opposite. But okay, but no uh So it's it's slow relative to the the degrees of freedom that move fast. So when we do a simulation of a molecule, we represent every degree of freedom that that we can with as high accuracy as we can, you know, at like a let's say classical level. So we're not treating any of the quantum.
SPEAKER_04So I just understand the complexity where we are.
SPEAKER_06Like how many degrees of freedom are there? Aaron Ross Powell Um So for a typical molecular dynamic simulation of something like a protein, there there might be order a million degrees of freedom.
SPEAKER_04And what are you and you're simulating you're simulating the protein doing what? Like moving?
SPEAKER_06Well, mostly it will be like wiggling and jiggling, right? So mostly it's wiggling around.
SPEAKER_04Uh so it's that awesome video of the like of DNA folding and unfolding, and it's like the the whole thing's like so crazy shaky. Yeah, yeah. It's like how does anything coherent happen with that much motion? You know, yet it happens.
SPEAKER_06Even those, I think to like the well-trained eye, over they'll look overly deterministic. I mean, everything is very, very highly stochastic at these scales. Um but you know, the things that we we would typically model would include you know all of the atoms of a protein as well as an environment of solvent and uh ions. Okay.
SPEAKER_04So you know And we're not even trying to get to like action of an enzyme. No, yeah. So this is this is just jiggling of a protein.
SPEAKER_06Just like a basic molecular dynamic simulation, you're gonna see some movement, maybe.
SPEAKER_04Which is where you get simple like uh signal transduction, that kind of thing.
SPEAKER_06Aaron Ross Powell Well, so that comes in at like a larger scale involving mini proteins.
SPEAKER_04But it's still like it's just a protein like moving. Like it's many proteins, but like if you're not a proteins interacting, right.
SPEAKER_06Yeah, yeah. Uh you know, typically doing something like catalyzing a reaction or phosphorylating something, right? And so you have these mechanisms by which uh physical interactions transmit some information from the extracellular environment down to a cellular decision. So uh so yeah, the number of degrees of freedom is incredibly large.
SPEAKER_04Aaron Ross Powell Like a million.
SPEAKER_06Yeah, or order a million.
SPEAKER_04Like we're like 10 to the six, that's where we are, something like that.
SPEAKER_06Yeah, it can be more, right? Like it can be much more than this.
SPEAKER_04But just for like a protein of like what size? Uh like how many atoms are we talking about? Like 10 atoms, like 50 atoms, 150 atoms.
SPEAKER_06So each residue of a protein will will be order like you know 10 to 20 atoms. So uh typical like a typical protein that we might do a simulation of would be anywhere between, let's say, 150 and 500 amino acids. Uh each of which is 20 molecules. Yeah. I mean 20 atoms. Yeah, order of magnitude larger than that in total in terms of total atom count. Right. But most of the degrees of freedom that we're actually you know putting computational effort into are solvent. Um so there's a lot of water around the protein. And like it it turns out that actually representing all of that solvent is is important for getting accurate results from the simulations.
SPEAKER_04And so it's not just water, right? Because it's like water with all kinds of like dissolved ions.
SPEAKER_06Yeah, water with ions, other stuff. Yeah. I mean, yeah, typically like you know, for like a a simulation in a simple setting, it would just be um physiological concentration of salt. So, you know, sodium chloride or uh or something, you know, potassium chloride at 150 millimolar concentration. Like that's roughly what the cellular ion uh concentration is. And so that's where we do most of the simulations. But um, but yeah, each one of these simulations has an enormous number of degrees of freedom. And the thing that really sets the time scale of how long we can simulate is uh the oscillation of hydrogens. So hydrogens are very, very light. They're bound to um you know atoms throughout the system. But they're also free-floating a lot.
SPEAKER_04So there's a lot of there's a lot of like acid-base, right? Like type of there's a lot of like hydroxyls in there that were the high where the Trevor Burrus, yeah.
SPEAKER_06So we in a classical simulation, we don't usually capture that complexity. There's um that seems like an oversight. Well, so that's a good that's a good question, a good I a good point. Um Did I already do that?
SPEAKER_05Oh no. No, no, no. You've got a million degrees of freedom, and it's like, but can't the high the hydrogen ions just like dissociate? You're like, well, actually we don't. That's one of the ones we know.
SPEAKER_06Well, I mean, so you can do it. And in fact, the guy down the hall, Tom Markland, is like a uh a world expert in how to represent protons, you know, free and solvent. And like the current generation of machine-learned interatomic potentials do allow for this sort of reactive chemistry to happen more fluidly. But the time scales associated with that acid-base chemistry are are typically kind of longer than we would would simulate. Or you can't because of the computation power, or because we can't, because of the computation power, it adds lots of degrees of freedom.
SPEAKER_04Um What's the time scale you're you're you're representing?
SPEAKER_06Uh okay, so yeah, I was just coming to that, right? So the the hydrogens are are oscillating super fast, right? The the time scale over which they oscillate is in the femtosecond regime, so 10 to the minus 15 seconds.
SPEAKER_03Okay, yeah, thank you. Yeah, I wouldn't have had the grade there.
SPEAKER_06So the time steps that we can take in when solving the equations of motion for a classical molecular dynamic simulation are on the order of femtoseconds. So we typically would use something between a two and five femtosecond time step, depending on various tricks that we can play. Um, and that sets the scale for how long you can simulate, because each one of these forward time steps requires that we compute all of the forces in the system, right? So that's like a pairwise interaction among a million degrees of freedom. And that has to then you know get propagated uh a single time step, and we have to do that over and over again.
SPEAKER_04So even to reach How do you even pick single time steps? Because all of these things are not like all these things aren't oscillating by a metronome. Well, like 400,000 of them could like shift sort of at the same time, 10 to the minus 16th, right? Yeah, 17th, and then the other like another 700 of them are gonna oscillate, like right?
SPEAKER_06Yeah, so so we, you know, everything evolves together, right? It's like it's like re real time, right? So everything everything changes together. We compute the interactions among everything. So that would be kind of continuous. Well, I mean it does basically. Yeah. So I mean there's discretization error that comes from it's like any time that you solve an equation, you know, that propagates something forward in time.
SPEAKER_04How did we get to Isenberg? That's so cool. That's so cool. Like we got to like, I I don't know, I can't know where it is and where it's going at the same time. Because there's discretization error.
SPEAKER_06Well, so there's discretization error, but I know it's not the same, like I know it's not actually like an uncertainty principle. It's not exactly the same, it's just because it's like this is this is like um numerical error. Like you think of it like it's not there's not a physical limit yet on how small a time step can be. Like we would have to go way, way smaller in time step before there's an actual hard physical limit. What's the limit? But uh well, it would be related to a plank length. It would write a course then. Yeah.
SPEAKER_04Um, and we're gonna get into string theory.
SPEAKER_06Like what the so the limit the limit for us is actually the opposite one. We we want to evolve these systems as fast as possible.
SPEAKER_04Yeah.
SPEAKER_06So we want to push the time step to be as high as possible without introducing too much error.
SPEAKER_04Aaron Ross Powell Because actual things that happen in a molecular system.
SPEAKER_06Interesting things in proteins are anywhere from the millisecond to second time step. No, I see what you're saying. I see what you're saying. It's like it's much longer. Right. So we need to do order, you know, millions to trillions to quadrillions of time steps in order to Then you can do two to five. Um we can do No, no, I mean we can do we can do time step, like you know, you can do hundreds of nanoseconds a day. So that's you know, millions of time steps a day.
SPEAKER_00Okay.
SPEAKER_06Um but it's still, you know, it's that's still limiting. Because on the time scale, like six proteins.
SPEAKER_04Six powers below where you're trying to get to with milliseconds, like right? Like you're still Yeah, yeah, yeah.
SPEAKER_06Yeah, so it's there's still a huge gap to milliseconds. And like the longest simulations that have ever been done, you know, that use explicit, like all atom molecular dynamics are in the millisecond range, but um, but those are quite expensive and for small proteins and done on special purpose computers, right? So so a lot of the work in in this field is trying to figure out smart ways of bypassing this sort of like brute force approach of just like set up the system, watch it evolve under, you know, time-dependent dynamics, and hope that something interesting happens.
SPEAKER_04Which is where the machine learning comes in.
SPEAKER_06That that is a lot of the direction that we're pursuing now is to use machine learning for these sorts of things.
SPEAKER_04Well, yeah, like something like that. To more effectively abstract, yeah, more correctly or something abstract out.
SPEAKER_06I would say a big part of the way that I think about using machine learning is that the the real power of neural networks in the context of computational physics is not really their like predictive capabilities, but it's rather just the generic utility of neural networks as ways of representing extremely high-dimensional functions. Okay. So, you know, in when we think about like running an MD simulation or sampling configurations of a molecule, this is really a question of how do we uh sample a very high-dimensional probability distribution, which is the probability of all of the states uh in a system. So those objects mathematically are extremely hard to represent because they're um they're so high dimensional. And so classical approaches like things called like kernel density estimation, which are kind of like old techniques for how you might represent high dimensionality.
SPEAKER_04You're gonna have strength, you're gonna have mathematical functions that are you know like thousands of thousands. It's just like yeah, it's like thousands of inputs, right?
SPEAKER_06So it's the they're they're really they're really difficult to represent for that reason, but neural networks offer a kind of uniquely well-suited parametrization of super high-dimensional functions. And like basically the whole, you know, the whole paradigm of generative modeling is built around the idea that we can actually just represent these high-dimensional functions. Like at all. Yeah, and and the whole and and it's working, right? I mean, like it's clear that you can do this. Like language models are representing distributions over time.
SPEAKER_04So there's another way we could have proven that you can do this, which is that our brains do it all the time, like our bodies do it constantly. Like that's what neural networks are modeled on, right? Yeah, yeah. Like yeah, so and like it's literally that it's like, how do you take, I mean, that's what your system is doing at all times, right? It's not like that's what I'd write, like attention, right? So like I do ADHD, and yeah, like that's my very specific field as I as I treat, as I I work with adults with ADHD. Yeah. And the construct of attention, like what I was interested in is what is consciousness, yeah, right. And as close as I can get is its attention, right? Um and then there's some more there, but uh I mean you're taking in, like, what are how much are you hearing? How much are you seeing? Yeah. Like how much are you like, and then there's all the things you're experiencing internally, right? And there's all the different ways the body brings that information to processing centers, right? And then has to like abstract out or just like remove what's not important, right? And then and then compute that, right? So it eventually becomes music. It eventually gets into it eventually, like it seems the best I can get is that it's harmony, that it's that like the the closest I can get to it is music.
SPEAKER_06Yeah. So I I don't I don't know about that. I don't know much about consciousness, but what I can tell you is that you do these these functions need to pay attention to a lot of degrees of freedom. Sounds like you kind of do know a lot about it. Sounds like, yeah, right. And uh yeah, so like, you know, paying attention to a lot of degrees of freedom is something that neural networks are really, really good at. And that that is something that is incredible. So around like 2018, 2019, when I started really like taking, you know, like bu building a lot of the toolkit around what got you into this? Yeah. So that's a good question.
SPEAKER_04So where does this where does this like where does this like catch you? Um like really catch you. Yeah. There's so much deep like same kind of thing. Like there's so many degrees of freedom. There's so many variables, right? But like there's all these different things you could be studying with your knowledge, right? Like why, why has this really got you?
SPEAKER_06Yeah, it's a it's a good question. Like I'm not really sure how how it happened myself in a lot of ways. I mean, so I, you know, I I studied math as an undergraduate. Like conscious and unconscious.
SPEAKER_01I love it, man. It's so it works. It's so all all the things that like operate at one level operate at the other level too. It's so great.
SPEAKER_06Yeah. I thought, I thought for a long time that I I just wanted to be a pure mathematician. Um and I went to the University of Chicago, which is like the least applied math-focused math department in the country. I mean, it's like no no applied mathematicians in the department at the time that I was there. Um, we didn't really like have courses that were very focused on applied or computation. All theoretical books. It was very, very theoretical. Yeah. There were a few subjects that I loved. I loved representation theory, which is really like an abstract algebra topic. Um, but around my junior year, I started working uh just with some biophysicists in a lab on a project. And I I, you know, I thought um I sort of started to appreciate that like a lot of what I saw as kind of you know the frontiers of science were really much more about understanding complexity and biological complexity was a kind of interesting um purview into that. And like I I realized at that time, uh also when I learned statistical mechanics, like that you know, this was a good scientific setting for trying to understand that class of property. Approach complexity. And in particular, that it had like a lot of mathematical richness, which I didn't previously appreciate.
SPEAKER_04Um isn't complexity theory its own thing?
SPEAKER_06It it is its own thing. It's like a little bit um, you know, I would say disjointed from the rest of physics, and that it has focused on dynamical systems a lot without really, you know, thinking about models of of chaos and and development of chaos. But like chaos is just a generic property of systems that evolve. Like um, you know, we we can't really anticipate you know the time scale on which something is going to decorrelate from its initial condition, but that that is a signature of chaos. And that's one of the reasons that sampling problems are hard, is that like we can't, you know, simply um steer things in the way that we expect. Um so yeah, I got attracted to these sorts of sampling problems and uh and decided to do a PhD in biophysics with Because you're still very far away from like running an actual experiment. Yeah, yeah, absolutely. Uh yeah, I mean everything that I do is computational. I mean, we we work a lot with experimentalists across a bunch of different domains, but um but it's definitely you know not uh not something that I directly do or have any expertise in how to do.
SPEAKER_04Like, well, but but your expertise is clearly very relevant to designing the experiment, to like what they choose to do. Yes, yeah, yeah.
SPEAKER_06And so and yeah, we try we try to work with them carefully to make sure that uh you know we're making accurate predictions about what they're gonna measure and you know, things like that.
SPEAKER_04But a lot of what you say is stuff like we can't, you know, we can't test it, right? You can't like there's this like what can we test? Like how can we which all sounds very experimental, but then but then right? There's this, yeah. Oh man, yeah. So much of what you say comes back to for me, it comes back to this there's this gap, right? Between layers of organization, or between um what we know and what we don't know, um, or what we can do and what we need to be able to do, right? There's this like right, yeah, which then but then also a word you draw like emergence, right?
SPEAKER_03Um trying to like, what is it? Something's landing for me here, man. I don't know what it is.
SPEAKER_06But I should so like Yeah, I mean I think the phenomenon that you're describing, I think, is a a really generic set of features of interesting research problems. Yeah, maybe when you know.
SPEAKER_05Maybe I can set something that is like profoundly like not interesting.
SPEAKER_06Well, that's awesome.
SPEAKER_02No, I don't I don't know, I don't know about that, but I just used big words to like describe something very straightforward.
SPEAKER_06It's just it's just that like, you know, in I I would say in the fields that I'm working in, there's a confluence of a set of interesting opportunities that have been enabled by the combination of like the physical theories and understanding, you know, reaching a level of maturity where we can formulate questions that are interesting, coupled with advances in computation and in machine learning that are allowing us to approach these questions with a level of computational sophistication that previously was not available to us. Um together, those two things are are kind of they're changing the problem landscape a little bit and they're pushing it more towards um, you know, questions that previously we just thought were off the table.
SPEAKER_04Is it okay if I if I try to land us somewhere specific? Yeah. What's a specific problem that we are now that like you were like, whoa, we could do that.
SPEAKER_06So so one I'll I'll tell you about that we've been working on a lot over the last two and a half years is uh the problem of understanding the mechanism of ATP hydrolysis.
SPEAKER_04Oh, do it, man. Do it.
SPEAKER_06Yeah, so so ATP hydrolysis is you know the chemical reaction that everyone knows about. Um most people learn about it in elementary biology, right? Like you you know, you learn that ATP is the uh the energy currency of the cell, right? So ATP is uh you know a molecule, adenosine triphosphate, that has three phosphate groups that are hanging off of it, and a lot of enzymes in the body will catalyze a reaction uh that hydrolyzes, that breaks off the gamma phosphate, so the third phosphate, and ATP goes to ADP.
SPEAKER_04Um could like you you don't have to explain. I mean you can just be like ATP hydrolysis. Okay, so what about ATP hydrology? Everyone knows this. So right. Like yeah, but this like sort of like of this podcast is kind of for people who like who are like they're like I'm I'm with it.
SPEAKER_06Yeah. So so this is uh you know this is one of the core reactions in in all of biology. The thing that we don't really know and has been hard to do.
SPEAKER_04Yeah. But fundamentally, it's the spark of life.
SPEAKER_06It is the thing that creates directionality in most biological systems.
SPEAKER_04That's a lovely way to say it. I would have said it's the fire that burns but doesn't consume. But keep but keep but keep going, man. Yeah, it gets I mean it gets recycled, right? Creates directionality. Okay.
SPEAKER_06It creates directionality. So like that's the way that I think about entropy and entropy production is that you know the the one entropy production.
SPEAKER_04Very cool.
SPEAKER_06Yeah, so the the one thing in physics that doesn't satisfy time reversal symmetry is the second law of thermodynamics, which says that entropy is an increasing function. The entropy of the universe is increasing function.
SPEAKER_04Forever increasing.
SPEAKER_06Um so the only thing in in physics that sets the directionality of time is the production of entropy. And ATP hydrolysis is a uh a mechanism by which we generate directionality by uh you know by producing entropy. Um so the way that biology leverages ATP hydrolysis is that it catalyzes this reaction. The reaction is incredibly rare in in solvents. So if you just take it in bulk, the time scale for it is quite long. That's make it made it very, very hard to study. Um so one of the things that has been enabled by um by using machine learning is building incredibly accurate representations of um the potential energy surfaces of molecules that uh, especially ones that are undergoing a reaction. Um so we have been leveraging all of the technology around uh how to build very, very good machine-learned interatomic potentials to study ATP hydrolysis in bulk. So just like you know, the basic question. And actually, you know, biochemically, this is not um very well settled in the literature. So there are there are a few different putative mechanisms by which the gamma phosphate dissociates.
SPEAKER_04Um we don't know how it's we there's debate.
SPEAKER_06There's debate. I think you could probably find people out there who would claim that we're so what are the possible what are the theories that are that we're not quite that have we haven't settled on one? So the two the two uh most widely discussed theories are that the mechanism is dissociative or associative.
SPEAKER_00Okay.
SPEAKER_06Um so man. Dissociative. Yeah. So dissociative would be basically that like you know, the bond starts to break first, and then you know, a magnesium ion that's coupled allows the hydrolysis to occur. Uh, the associative is more focused on uh the uh coordinating water uh going to the magnesium first. So these two different settings you know lead to different reaction rates, they lead to different barrier heights, they lead to different reaction energies. So, like, you know, what's the difference between the initial energy and the final energy?
SPEAKER_04So Which would make me think we would have been able to test that.
SPEAKER_06So there's debate where you know simulations uh which uh have been done using uh you know like pre-machine learning methods, uh, and experiments do not agree uh on this. And uh there's a lot of debate about you know who is right.
SPEAKER_04Um I I sort of tend to Wouldn't we be able to just tell by the reactions that they catalyze in in a living system? The amount of energy.
SPEAKER_06So this is the uncatalyzed setting, right? So this is just the this is basically just asking what is the dissociation like in water? And in that setting, it's actually really hard to measure, like it's hard to do the experiments because the reaction rate is low.
SPEAKER_04Trevor Burrus, Jr. Why is it important to know how it happens outside of life?
SPEAKER_06Well doing it. That gives you a lot of insight into how the catalysis works as well. I mean, uh our longer term vision is to study it in a lot of protein systems, but we want to nail down the reaction mechanism in the bulk first, just so we understand how this happens, right? Like what is the process by which the gamma phosphate breaks off? Trevor Burrus, Jr.
SPEAKER_04But you could be wrong, right? Like you could discover that Well, one can always be wrong in the other thing. You could discover that like that that in the bulk, yeah. Like it's definitely the water molecule is associating first, and like that's what's driving it, right? Yeah. Um But that I mean that's an answer to a specific question, right? And then in in actual like a phosphorylase, right, like it might be exact, it might be the other thing. Right?
SPEAKER_06It might be, but like we'll have we'll have a good toolkit to study that because we'll build a representation of this reaction pathway that allows us to really nail down the electronic structure at intermediate points, and we'll build a very good representation of the potential energy surface. So like all of this scaffold that we're doing in the bulk case, we get to leverage in other settings. So having a good representation of what the underlying physics is like is something that's highly transferable. So when we go and we take that representation and then we put it inside a protein, like we can sort of re-optimize and you know study the problem in situ in a protein in a different way. So we'll get different results there.
SPEAKER_04Do you have a sense for the for the pathway of the electrons, right? Like for the how the potential energy is shifting through the system um for each of the different theories. Where like if it's this theory, then this is the whole thing. And if it's this theory, then this is the whole thing. That we just need to figure out which one is accurate?
SPEAKER_06Uh so so we do. We've we've studied the pathways in all in uh a few different putative settings. Um I I don't want to like I don't know exactly that we have finalized our our understanding of our results.
SPEAKER_04Yeah, sorry if I'm like far afield from your specific.
SPEAKER_06No, no, no. It's it's just that like I don't I don't want to come out and say like, oh, it's definitely concerted. Uh and that is kind of like directionally where the results point, but there's still there's like more follow-up to do so that we are confident that we're we're saying exactly the right thing about how the uh about the mechanism.
SPEAKER_00Yeah.
SPEAKER_06Um the implications of of this are uh you know, are more that we're we're we will have built a representation of the potential energy surface that we can now embed in proteins. And so we can study modulation of reaction rates by different protein environments. And like that's the that's the compelling part in my mind. So that that's uh that's something that you know dictates.
SPEAKER_04Then you you expect that different different environments will actually do this very basic reaction differently, or Mike.
SPEAKER_06Very much so, and and with with different rates also. Like so you know, there's a lot of competition for ATP uh hydrolysis, right? Like things do it uh slowly, quickly, right, and you know, with different purposes.
SPEAKER_04So this is sort of news to me. So there's many enzymes that that catalyze this reaction.
SPEAKER_06Oh yeah, tons, tons. So um you know, the probably the most famous is ATP synthase, which is a very particular yeah.
SPEAKER_04Oh, but it's the reverse also.
SPEAKER_06Yes, yeah. So it does two things, which is uh kind of rare in enzymes. Um so it both acts as a proton pump uh by catalyzing a reaction. Tell me what ATP synthesis gets mad. Uh yeah, okay. So so yeah, so you know all about this, right?
SPEAKER_04Do you think apparently not like okay, but oh right.
SPEAKER_06But also, you know, there are things like um kinases, like RAS, for example, that finds a GTP, which associates.
SPEAKER_04Which is where that's your signal transduction.
SPEAKER_06Yeah, so so signal transduction is driven by these types of reactions as well. Um molecular motors typically rely on GTP or ATP um uh again as a source of reverse of irreversibility. Um so the way that they you know are rectifying motion.
SPEAKER_04Well, you see like you mean like a flagellum. Like when you say molecular motor, but that's obviously that's not molecular.
SPEAKER_06Molecular motor is like at this uh single molecule scale, there are proteins that act sort of as motors that so ATP synthase is one that often gets brought up as a molecular motor and that it's like acting as a pump by rotating. It's like a rotary motor. Um but for example, uh kinesin and dynine are um proteins that walk along microtubules and carry carbon.
SPEAKER_00That's what I was yeah, yeah.
SPEAKER_06Um and so so the way that they work is that uh they hydrolyze ATP that leads to uh unbinding from so that you know they typically have two legs or multiple legs, right? And they'll take steps where with each unbinding. With each unbinding.
SPEAKER_04Yeah, yeah.
SPEAKER_06And there's a directional rebinding. So you know they move along this polarized filament in in with higher probability in one direction than the other. But the thing that really you know sets the cadence of these motors is the hydrolysis and r uh of ATP and then rebinding ATP. So each cycle of this does, you know, I think when we say motor, we're tempted to think that what is happening is that the energy that comes from hydrolyzing ATP is somehow being put into the system and that's pushing it to do something. But at like rocket fuel. Like like rocket.
SPEAKER_04Like literally pushing.
SPEAKER_06Yeah, but but at the nanoscale, that's not that it's not possible for that to happen. The Reynolds number is too low. And it's just there's no such thing as a ballistic force at the nanoscale. Like you you can only the only thing that you can do is act as a ratchet. You can only fluctuate and rectify fluctuations. Like that's a a very, very important organizing principle of like how biology does work at the nanoscale. Yeah.
SPEAKER_04It's all ratchets. I mean ATP synthetase is very much a ratchet.
SPEAKER_06Yes, everything, everything has to operate effectively as a ratchet because things are highly stochastic. And if you you know, in a at a very low remote.
SPEAKER_04Define stochastic here now for me. Uh like why does that make it have to be a ratchet? I mean, it may it would make sense that like it could only be a ratchet if ballistic force isn't available.
SPEAKER_06So so another another way of thinking about it is that in in these environments, the um ballistic, the magnitude of ballistic forces compared to diffusion is extremely small. So if I push on something, it's gonna just come back. Right. Right. Like there's no there's no continued momentum in one direction. Yeah things just fluctuate. Yeah. And so so the only thing that you can really do to get directional motion is you can put up a wall behind something when it fluctuates. So it moves and then you move the wall and then it can't move back anymore. Okay. And so like that kind of ratcheting mechanism is is basically the the the way that all of these molecular motors tend to work. Um and you know, the ATP hydrolysis basically acts as you know an event that allows you to, you know, initiate one cycle of the ratchet. Yeah. Um, and so this continued rebinding of ATP.
SPEAKER_04So the process of the enzyme actually doing the hydrolysis and and and then releasing and rebinding is the thing that's gonna determine how the ratchet can the speed the ratchet can operate. Exactly. Yeah. Yeah. And that's all gonna then come down to molecular like atom like this degrees of freedom, like this like very complex, like how does the unbinding and rebinding and with the solvent around and right, like all this stuff.
SPEAKER_06No, it's uh I think the one of the the most interesting parts about biophysical dynamics is that we have no idea at what length scale we can stop caring. Um you know, we we don't really have the ability to robustly coarse-grain, you know, to large length scales uh in any way that seems to be universally effective.
SPEAKER_04Same's true in psychiatry. Yeah. It's exactly the same thing. It's like we don't know at like okay, at what at what point can we ignore the body? Yeah, right? At which parts of the body? Yeah. Right. Like, do we have to be paying attention to molecules? Well, now we have meds that are neurotransmitter molecules, right? Like, okay, but then but then also how does that propagate up through like 20 different layers of system to how a human experiences?
SPEAKER_06Right, right. Yeah, I mean, I think yeah, you could say that uh psychology is uh is a biophysics problem in a way, right?
SPEAKER_04Um not in a way, like very concretely.
SPEAKER_06Yeah, uh I think you know, I I think it's it's interesting that like even you know, for predicting protein motions at long time scales, like we can't get rid of the hydrogens very well. Like this is not something that has been robustly done. So like we can't get rid of the solvent, we can't get rid of the hydrogens. Like every degree of freedom matters a little bit in ways that you know we haven't been able to um build highly accurate theories that actually just integrate out that information.
SPEAKER_04If you could find what you're looking for, what would it be? Uh and I'm not saying like you know what you like that there's like a specific answer, but like if there is like a specific question that arriving at an answer, you'd be like, I got it.
SPEAKER_06So I think one of the questions that has driven me a lot is trying to understand how organization emerges in systems that are driven very far from equilibrium. We tend to think of like you know, the applying of like putting in energy and like applying forces as being things that l often lead to disorganization, that they, you know, they sort of push things, they make things, they make the dynamics more chaotic. But in biology, it's really the opposite. Typically, you know, energy and force transduction is used as an organizational principle. Um, and it's very hard to see how that dynamic spontaneously has arisen through the evolution of these systems. So the dynamic of harnessing the fire instead of being burned by it. Something like that. Yeah. I mean, I I think I think that's an interesting way of putting it because that that is like, you know, that is what these systems have spontaneously learned to do. That you know, they have been driven into building increasingly complex organizational structures. Trevor Burrus, Jr.
SPEAKER_04Yeah, there is a sense of driving towards complexity. Yeah. Which there's a sense of like that's the aim of life, is to like continuously achieve mastery of higher and higher levels of organization. Like that's just where we're going.
SPEAKER_06Yeah. Like from from molecules, whether or not it's intentional, I have no idea, right? But it's certainly something that that seems to have happened. I'm not sure it matters. Right. Like I'm not sure it's not. This is true across different length scales, where you know energy transduction is just happening all the time. You know, it's it's part, it's part of modeling these. Systems and understanding them. And it's a piece of information that we we often don't think about when we think about doing biophysics. Like a lot of biophysics is focused on what are the structures of things. And it takes a very static picture of how this actually operates and works. And like, you know, there's a huge amount of investment right now into technology for designing things like binders, like antibodies and drugs. And that also is driven very much by a static experimental picture of like here are the crystal structures that we know. This is what binds here. But in fact, like all of these objects are highly dynamic. They're all interacting with one another. You know, achieving a high degree of specificity really requires paying close attention to the underlying dynamics and thermodynamics of these systems.
SPEAKER_04So the question you're asking is why does it move?
SPEAKER_06Well, I think we know why it moves, right? I mean, we we know why things move at some at some basic level, which is that like there's heat, right?
SPEAKER_04The heat is caused by you just specifically said ballistics doesn't work at this scale.
SPEAKER_06Aaron Powell Well, the but the heat just causes fluctuation, right? Like things move because there's like, you know, there's energy in being coupled to the environment. So like we are at some ambient temperature where you know things fluctuate because there are thermal fluctuations. And then on top of that, like the systems that are living in this thermal bath also are exchanging energy with one another and exchanging, you know, we we kind kind of compartmentalize things into like the scale of like a cell. Well, uh, you know, the cell itself as a compartment is something that's exchanging various forms of energy with its environment. It's exchanging heat, it's exchanging metabolites.
SPEAKER_04I can clarify the question. Yeah. Why does it move in the direction that it moves? Not why does it like move randomly? Yeah. But like why does life move this way?
SPEAKER_06Yeah. I I don't think we have a better answer to that than the second law appears to be directional, right?
SPEAKER_04Like, yeah, but but life seems to be the one thing that operates against it.
SPEAKER_06Well, it it doesn't because you know the caveat with the second law is that you can use energy to reduce it.
SPEAKER_04Like I know like globally it doesn't. Like if you take into account in like the waste product, right? Like, yes, system as a whole still, like the the the law still holds, but life seems to like here the universe moves towards disorganization and life seems to move towards organization.
SPEAKER_06Yeah, it's more disorganization outside. But I mean it's using energy, right? It's using energy to counteract the effect of of entropy.
SPEAKER_04Yeah, so why does it move in this direction? Like what got yeah.
SPEAKER_06Well, there's there's just enough energy around, I suppose, right? It's like, you know, we're we're exploiting the energy that is available.
SPEAKER_04No, but it sounds like this is the question you're asking. Explain the but like but like why? Like why exploit the energy that's available to march down this, to march down this microtubule like this direction? Like what is it, how did life decide that that's the same thing.
SPEAKER_06Yeah, so I think to ratchet this way. Yeah, I'm really hesitant to ask the question why, because I think that why implies intent.
SPEAKER_04But yeah, I think trying to I'm saying it, I'm trying to say it without intent. Yeah, but yes, but I agree with you.
SPEAKER_06So so I think the the answer to this question it from my perspective or the the the way of framing it is more um what are the mechanisms by which these types of structures emerge from the input of energy? Right. It it's a is a really just the morphine of the thing. Have you been able to answer that in any stream in any So I think in specific systems we understand how applying force, how inputting energy modulate the dynamics in ways that affect um mesoscopic properties, so things that look much larger length scales than the individual components. Um in general, I I would say we we still don't really have a lot of people.
SPEAKER_04And it's just like hand waving, like because the lower scale does what it does, and we can't we can't figure it out. So we just know that it does. So let's start from here where we can measure and we'll we'll figure out stuff at this at this time scale, at this length scale, right?
SPEAKER_06Right. I think I think that that's like you know, back to the point that I was making earlier, I think that that's a one of the interesting aspects of this uh field. It's simply the case that we don't know at what length scales we can stop paying attention.
SPEAKER_04Yeah.
SPEAKER_06Um, it seems like we we need to really attend to every length scale in order to make the most accurate predictions. There's there's some cases where we can build really minimal models that capture kind of universal phenomena in systems, but that approaches things from uh the opposite direction, right? And it's not it's not as much of a molecular theory of how things work. It's much more of a like macroscopic theory. Yeah, these are the these are the ingredients, these are the way that these ingredients act interact at a minimal level. And then you can sometimes make predictions that are directionally correct or capture some of the universal behavior. Right. But when you want to understand, you know, specific modulation in systems, you know, what happens if you mutate this residue of a protein and why does that lead to a disease, whereas the other thing doesn't? Like those are questions that involve so many length scales that it's extremely hard to answer them with a theory that only pays attention to a small um dynamic range.
SPEAKER_04Why'd you pick the length scale you picked?
SPEAKER_06The length scale of like a mutation of a protein.
SPEAKER_04Aaron Powell No, I mean the length scale that you're assessing, the or the time scale. Like why'd you pick femtoseconds? Why'd you why'd you pick why'd you pick like an it like individual like where is the hydrogen ion actually shifting?
SPEAKER_06So I would argue that we haven't picked a length scale. Like I think. So where you're actually trying to investigate. Aaron Ross Powell The range over which we are looking, I would say, is bigger than um than what you would traditionally see in theoretical chemistry, which you know, theoretical chemistry tends to focus very much on the atomistic resolution.
SPEAKER_00Okay.
SPEAKER_06And we have really tried to be more expansive than that. So like I'm interested in understanding how molecular degrees of freedom, so things at atomistic resolution, inform properties in non-equilibrium systems at the length scale of hundreds or thousands of proteins. And to get to those length scales, like things that involve self-assembled collections of proteins, we do really need to um reduce the dimensionality through some sort of coarse graining.
SPEAKER_04Right. Um and so you're starting at the small like at the small scale you can to figure out what how can we coarse grain like here so that we can then very much.
SPEAKER_06And uh the the thing that's difficult about this, you know, connecting these different length scales is that often the molecular degrees of freedom matter a lot. So, you know, the choice of mutating a residue is not like a not a random choice. Like that's a that's a type of problem that biophysics cares about a lot. We're very, very interested in biophysics generally on how protein function is related to sequence. And you know, that's extremely sensitive to even small perturbations in the sequence a lot of the time. Yeah. So you know, there are diseases that are caused by point mutants that lead to either misfolding of a protein that inhibits its activity in one way or another. There are diseases that emerge from the assembly of proteins when you know something uh gets expressed too many times. For example, like Huntington's disease is caused by self-assembly of the But this question of length scale, this question of like it's the communication across length scales.
SPEAKER_04It's the like how does one length scale like inform the other?
SPEAKER_06Exactly. Yeah. So like, you know, we can't look at the macroscopic system. Right. We can't look at the macroscopic outcomes and say, you know, this arose because there was a difference in the molecular scale. Um but we would like to be able to say things like that. You know, we would like to be able to look kind of phenotypically at systems. Not very often.
SPEAKER_04Not very often. I mean this this um this diphosphate did not bind to the site in the ATP synthase that will accept the one phosphate that's going to attach to the ATP, right? Because it's the wrong size. Right? So like it didn't bind.
SPEAKER_06Aaron Powell, so that that's a very molecular answer, right? And I think that that's something that looks at molecules, like the only way you would know that is you measure things at molecular scales. Yeah.
SPEAKER_04And you wouldn't be saying this is why. You'd just be saying you'd just be saying it just doesn't.
SPEAKER_06Right. I mean, the the readouts that we often have are like, you know, this cell looks healthy, this one doesn't. Right. Right. Or you you have some you know fluorescent measurement of you know where something is in the cell. Like this protein is over here, it's not over here. Yeah, yeah. But we don't necessarily know a priori, you know, what the molecular scale phenomena that led to these different self-organizations. Uh we don't know how to connect these things very well.
SPEAKER_04And the and the things that we need to understand are like how how do the hydrogen ions move?
SPEAKER_06How do the We need to understand how everything moves, right? I mean, we just we need we need much better understanding of the dynamics.
SPEAKER_04We don't understand how the molecules move, right? Because that's a different scale. We don't need to understand how the how the um how the neutrons and protons. Yeah. So like it doesn't have to be at strong force, weak force. Like we don't have to, right?
SPEAKER_06Like there is all that I'm saying is that these things are are very strongly coupled across length scales. Right. So the way that the molecules move informs the way that the materials assemble, which is the same thing. Top down and their properties. Yeah. Right. So, you know, w when we do a rheology measurement of some elastic network of proteins, like changes in the individual amino acids, you know, even sometimes protonation of certain amino acids can affect the material properties at a large length scale.
SPEAKER_04So if you want to look at how a kinase is pulling off a phosphate to run a ratchet, right? What are the things you have to look at?
SPEAKER_06Like it what are the what are look what are the yeah, like what's the Yeah, again, so like I I think that that's like a that's a molecular question for which you know we could devise a strategy. Like this is something that we're actually quite quite interested in kinase function and uh kinases binding things with. Yeah. So let's do it. Let's do it. So so that's something that that really uh you know to me involves b basically just molecular length scales, right? So it's it's really something that you would try to answer by doing you know targeted simulations using free energy methods to really explore this at the length scale of like what is the mechanism of this catalysis? Right. That like that's something that that we could really view at an atomistic length scale. The time scales are long.
SPEAKER_00Yeah.
SPEAKER_06So there we need to use uh statistical sampling tricks like important sampling. But but the length scales are all kind of, you know, uh they're they're sort of fixed, right? It's it's at the length scale of, you know, this is one kinase and it's like we're we're somewhere between angstroms and nanometers. Yes. Yeah, yeah. So it's like, you know, tens of nanometers is like the maximum length scale at which you would do everything atomistically. But uh I think a lot of the more interesting questions are sort of ones that span multiple length scales. So like one of the projects that we've been working a lot on with uh my collaborator, Alex Dunn, is thinking about the way that um muscle tissue self-assembles. Do it. And so the sarcomere is kind of the organizational unit of muscle tissue. And that's something that involves hundreds of actins, hundreds of myosclines. Lots of ratchets. Yeah, lots of rather forces. Um I mean, there are dozens of proteins, uh dozens of protein types involved in these uh these macroscopic assemblies.
SPEAKER_00Okay.
SPEAKER_06And you know, it's very clear that small changes, small mutations can lead to disease states in this type of tissue. So the myosin, you know, can uh can really impact the large-scale self-organization, um, even with you know little mutations. Right.
SPEAKER_04There's a lot of dystrophy stuff here, yeah.
SPEAKER_06Yes. And so so all all of that is kind of connected through the molecular length scale to this emergent structure that exists on the length scale of thousands, hundreds of thousands of proteins. And we don't have good ways currently of saying, I'm gonna make this mutant in a myosin. What's the impact on the structure that I get? How is it gonna affect the dynamic function of that?
SPEAKER_04You're trying to do the mathematics of emergence.
SPEAKER_06Well, we're at least trying to find ways of simulating, you know, and understanding how these small land scales are. Yeah, I mean, like that's what you'd like. Trevor Burrus, yeah. I mean, I would love to do that. It's really cool, man. It's really, really cool. I would love if I if I felt bold enough to say something like that. I mean, I think that's what you just said.
SPEAKER_01I just like said it like, right?
SPEAKER_04I I think you're constraining it, right? You're because you could just you could like, okay, but then there's also like the emergence that arises from all of these cells acting together. And then there's emergence that arises from all these muscles acting together. Right. Like, okay, right. It's it's like from every length scale to uh like so then there would this okay.
SPEAKER_06So then my question is Well, I I mean if you take the definition of emergence to be w how are length scales connected to one another, then I think very much we're trying to do that. Yeah. Um we're we're trying to understand how dynamics and spatial scales, you know, couple across many orders of magnitude.
SPEAKER_04Right, but so let's narrow it down, right? So like what what is the number of orders of magnitude at which you start to see emergence? Right? At which you start to see behavior that can no longer be clearly understood just as um the sum of the things that are happening at like is it is it a hundred? Is it a thousand? Is it ten thousand? Right? Like what because that if you could answer like that like that question alone, yeah, it'd be like and and there would be different kinds of emergence, right? Like there's different, right? But but m you you gotta constrain it to be able to like what's the smallest like length scale like uh gradation, I'm not sure the word I'm looking for there, but like what's the what's the smallest amount of change in length scale in which you can see something which you can clearly say is that's emergence.
SPEAKER_06So so what do you what do you take the definition of emergence to be?
SPEAKER_04What I just said, that the the uh the whole is greater than some of its parts. But but more comp more concretely, that that there is a behavior happening at uh at a certain length scale that cannot be um understood clearly as just the simple interaction that are happening at the lower length scale, right? This is where earlier I said like when it comes to con like when it comes to consciousness, the closest I can come is is music, it's harmony, right? The closest we can get to um what is human consciousness at our current ability to assess that um in a living system is coactivation of different areas of the brain, right? The stuff you can see in an FMRI where you see like communities of activation, right? And we don't even really know based on the math whether like activation means they're both actually active. It's just like they're both sending the same, they're both a positive signal that we're thresholding in a certain way, right? Like it's the like I've talked to people who do that math, and it's like there's so many layers to that math, and like we don't really know what we're actually saying is happening, but we do start to see things arise. Like you can look at an entire brain a priori with these tools and it and see the different regions that you already know do things like motor strip and do like they will they will they you will be able to clearly see how they are activated together, right? Like they will they will show up as coherent communities. Right. And then we get into things like the um the default mode network and the ventral attention network, which like are different conscious experiences. Um the default mode network, are you familiar with this stuff?
SPEAKER_06Not not at all. No. Okay. Oh man, dude. Uh dude, the default mode network. I told you, I know nothing about consciousness.
SPEAKER_04Oh, you would love this stuff, man. The default mode network is um just a uh a predictable set of brain regions that activate, right? Like whatever, that are like coactive when you're when you're awake but not doing anything. It's essentially the daydreaming mind, right? Okay. Um and then if you then focus on something, different brain regions like sort of like quiet down, and other brain regions wake up and you get the ventral attention network, and now we're doing something, right? And there's a bunch of differences, a salience network, there's like different ones, right? Okay. Um but what's really going on there? Like, how is that consciousness? And as close as I can figure it, it's the it's it's I'll give you okay, um harmony.
SPEAKER_00Yeah, right.
SPEAKER_04Where that honestly, that may be the maybe I already answered the question. What's the what's the smallest length scale at which something emergent arises? Well, it's two voices, right? Where you have one sound and you have another sound, it's two notes. And together they make harmony, right? Now maybe you could you could maybe that's not a good one because you can just do the math and be like, well, that's because it's a fifth of which is which is this, this overtone, right? And we can see how the you know how they're like why that's a specific and a different overlapping of of the wavelengths than than you know, something that's a slightly off. So it's not a harmony, right?
SPEAKER_06I mean, if I narrow the definition to like what is the way in which function arises from components in molecular systems that on their own don't do anything, like you know, then you can kind of narrow you can narrow down to you know positive feedback loops, you know, the types of things that do arise in signaling networks. Like there there is a sort of like emergent behavior in the sense that you know these components in isolation or without the input of energy or external forces that they they would be stationary.
SPEAKER_04We've already got it. Like ATP like ATP doesn't the phosphate doesn't hydrolyze off all that regularly.
SPEAKER_06So like certainly doesn't ratchet. But but I think the the example of sarcomeres and like the assembly of muscle tissue is kind of an interesting one where you get like a really you know large length scale functional unit arising from molecular pieces, right? And like the length scale over which this looks like emergent behavior, I'm not sure exactly where you would put it because like I'm curious where you would put it. Well, so I mean, you know, there's self-assembl like you know, the myosin makes a thick filament. Like the the thick filament itself is an assembly of proteins. The actin is a self-assembled protein, it's a big polymer, right? And you know, these things are all interacting with one another in a way that positions them so that they can respond, you know, to and contract, right? And to actually like you know, make an active muscle. Right.
SPEAKER_04Um, which at the smallest which doesn't need to be a whole muscle. I mean, it could be it could be uh something like a flagellum on a on a you know on a right. Yeah, I mean but it's still gonna do like it's it's something that's gonna be achieved some purpose that life has decided, whatever intentional, right, uh has decided is worth expending energy to do because probably it uh achieves obtaining more energy. Yeah, right? Like let's pump this flagellum because it gets us to that food over there, right? Like right. Um that like for some reason life has decided this is uh a valuable expenditure of energy, right? Probably because it achieves the input of more energy, right? But not too much, right? Like not more than the system can handle, right? The just like the right amount of energy, right? Um that can be taken in, processed, whatever, right? Um but yes, okay, so what's the smallest scale at which um you could see intentional activity, right? Is it like how many, how long uh an act and filament you need? Like how long a mice and filament do you need? Do you need both, right? Maybe, right? Or do can you do it with just one? Right? Like let's like where let's limit the like and let's get down to the place where this is the smallest, we can say this is the smallest scale at which um an action that seems to have a purpose is arises.
SPEAKER_06I think this is really about the way that we formulate the, you know, this is really more about the perspective of you know the person watching the system really thanks.
SPEAKER_04I'm not sure that's maybe. So you can go down like So how do you constrain the the the no the length like how come you're not also worried about meters? How come right? How come you're not worried about like people walking? Right? Like how come you're like how come you're not worried about multiple cells interacting where you are, right? Like that's when it's like how do you constrain like the the the width of the of the ruler that you're concerned about? Well Or interested in.
SPEAKER_06I think the you know the answer is partially that like I feel like I have a toolkit to study things on the range from like the atomic scale to the meso scale. Like the math you have. Yes, yeah. So like the the mathematical toolkit that I have is appropriate for for those sorts of coupled length scales, right? And like that's something you know that I think of as fundamentally being about like a particle point of view, um, which is also incidentally like how a lot of the machine learning theory that we do is formulated, right? It's like thinking about the parameters as particles and you know, using m the mathematics that we use for these types of systems to describe those interacting systems.
SPEAKER_04Oh, so interesting.
SPEAKER_06Um assumptions that coincide.
SPEAKER_04But please, yeah.
SPEAKER_06So so I would say that like, you know, I I'm a little bit constrained by like technical abilities that are you know, tailored to a certain set of length scales. Um but at even at the smallest length scales, I think you can see things where there are uh you know the systems have evolved to do things that we think of as being like actions. Purposeful or something. Yeah, that that achieve something. Yeah, it ratchets. Yeah, it does clearly some sort of some sort of transformation of the physical system that you know leads to entropy production, which gets rectified into directional motions.
SPEAKER_04So that might be the thing. It's like and so like it's back to how does how does the kinase that actions the ratchet. Like what's the right like how does that that's that's the the range.
SPEAKER_06Yeah. Unfortunately I don't think that there's like a limit as as to like what the smallest length scale at which you can like No you definitely said that. Well yeah yeah well no which is great. Yeah. Right. Yeah. So I I mean you can just go deeper and deeper and you you can start you know focusing on the electronic dynamics, right? Yeah. You can you can always go you know to a higher level to try to understand like where where this you know where this behavior starts. I think for chemistry, you know, we normally don't go beyond the electrons, right? Like this is, you know, we we we are we are in a regime where things like you know the things that arise in high energy physics are not relevant because we're not at those energy scales. It's different math.
SPEAKER_04But like you have to get into quantum mechanics and it's different math.
SPEAKER_06Well so we do tons of quantum mechanics. I mean the chemistry department is full of people who primarily do quantum mechanics and I mean like obviously like globally, right. But like but but in your work. We we do a very small amount but like all you know all my colleagues are are you know are are doing quantum mechanics all the time, right? It's just that you know we we are interested in energy scales that are relevant for chemistry, which means that like we're interested in like a temperature scale from ambient conditions to like you know a little bit below to you know maybe a couple thousand Kelvin. Yeah like we're not going to see nuclear fission and fusion because we're not at you know 10 million Kelvin. Yeah yeah um so so it's just you know it's not uh an operative set of questions like those things we can treat like atoms we can treat as pretty staple at you know at the temperatures that we operate on on Earth.
SPEAKER_04Yeah yeah in living systems. Yeah yeah yeah yeah and this is cool yeah this is really cool thanks yeah it seems and I like there's like you're a pretty successful guy.
SPEAKER_06I mean that's what they tell me just get started yeah but like but like you know professor at Stanford's like that's not too that's not too shabby not too like leaving to go do like pre-cool gig like not too shabby right yeah um and if I could sort of distill the wisdom I hear from you right is that you use the tools you have partially yes yeah I mean we're constrained by you know some set of tools and like don't don't don't worry about the tools you don't have man like the tools you have are enough like use the tools you have yeah you can do interesting yeah you can do interesting stuff you don't want to be tool too tool bound right like it's uh you know it's good to learn new things also but um but yeah definitely like you know there's some sort of core set of approaches that you can leverage in a lot of different contexts uh that's that's kind of nice uh I mean that's one of the nice things about um about statistical mechanics and physics statistical physics is like you you get to work on a a pretty diverse set of problems with a shared set of tools.
SPEAKER_03Yeah.
SPEAKER_06Yeah. Yeah.
SPEAKER_04Well and you and you get to find cool people.
SPEAKER_03Yeah absolutely man this was a jam yeah yeah thanks for doing