Dmitry Korkin: Evolution of Proteins, Viruses, Life, and AI | Lex Fridman Podcast #153

Lex Fridman 0:00 The following is a conversation with Dmitry Korkin, his second time on the podcast. He's a professor of bioinformatics and computational biology at WPI, where he specializes in bioinformatics of complex disease, computational genomics, systems biology, and biomedical data analytics. He loves biology, he loves computing, plus he is Russian and recites a poem in Russian at the end of the podcast. What else could you possibly ask for in this world? Quick mention of our sponsors: Brave Browser, NetSuite business management software, Magic Spoon low carb cereal, and Eight Sleep self-cooling mattress. So the choice is browsing privacy, business success, healthy diet, or comfortable sleep. Choose wisely, my friends, and if you wish, click the sponsor links below to get a discount and to support this podcast. As a side note, let me say that to me, the scientists that did the best apolitical, impactful, brilliant work of 2020 are the biologists who study viruses without an agenda, without much sleep, to be honest, just a pure passion for scientific discovery and exploration of the mysteries within viruses. Viruses are both terrifying and beautiful. Terrifying because they can threaten the fabric of human civilization, both biological and psychological. Beautiful because they give us insights into the nature of life on Earth, and perhaps even extraterrestrial life of the not-so-intelligent variety that might meet us one day as we explore the habitable planets and moons in our universe. If you enjoy this thing, subscribe on YouTube, review it on Apple Podcasts, follow on Spotify, support on Patreon, or connect with me on Twitter @LexFridman. And now, here's my conversation with Dmitry Korkin.

Lex Fridman 1:58 It's often said that proteins and, uh, the amino acid residues that make them up are the building blocks of life. Do you think of proteins in this way, as the, uh, basic building blocks of life?

Dmitry Korkin 2:11 Yes and no. So the proteins indeed is the, the basic unit, biological unit, that carries out, uh, important function of the cell. However, through studying the proteins, and comparing the proteins across different species, across dis- different kingdoms, you realize that, uh, proteins are actually a more, a much more complicated, uh, so they have, um, so-called modular complexity. And so, uh, what I mean by that is, um, an average protein consists of, um, of several structural units. So we call them, uh, protein domains. And so you can imagine a protein as a string of beads, where each bead is a protein domain. And, uh, you know, in the past 20 years, scientists have been studying, uh, the nature of the protein domains, 'cause, uh, we realized that it's, it's, it's the unit. Because if you look at the functions, right? So, so, uh, many proteins have more than one function, and those, uh, protein functions, uh, are often carried out by those protein domains. So, um, we also see that, uh, in the evolution, those proteins' domains get shuffled. So, so they act actually as, as a unit. Also from the structural perspective, right? So, you know, y- uh, some people think of, uh, a protein as a sort of a globular, um, molecule, but as a matter of fact, is, is, uh, the globular part of this protein is a protein domain. So we, we often have this, uh, you know, again, the, the, the, uh, collection of these protein domains, um, align, uh, on a string as beads.

Lex Fridman 4:14 And the, uh, the protein domains are made up of amino acid residues, so which are-

Dmitry Korkin 4:18 Yeah, so it's, it's, it's-

Lex Fridman 4:19 So this is the basic buil- so you're saying the protein domain is the basic building block of the function that we think about proteins doing. So, of course, you can always talk about different building blocks. It's turtles all the way down. But it's, uh, there's a point where there is, uh, at the point of the hierarchy where it's the most, the cleanest element block based on which you can put them together in different kinds of ways to form complex function, and you're saying protein domains. Why is that not talked about as often i- in popular culture?

Dmitry Korkin 4:55 Well, you know, there are several perspectives on this, um, and, uh, o- one of course is a historical perspective, right? So, uh, historically, uh, scientists have been able to structurally resolved to obtain the 3D coordinates of, uh, a protein for, uh, you know, for smaller proteins. And smaller proteins tend to be a single domain protein, so we have-

Lex Fridman 5:21 Got it.

Dmitry Korkin 5:21 ... a protein equal to a protein domain. And so, so because of that, the initial suspicion was that the, the, the, the proteins are, they have globular shapes, and the more, uh, of smaller proteins y- you obtain structurally, the more you were, you became convinced that that's, that's the, the case. And only later when, uh, we ha- we- we started having, um, you know, uh, a- alternative approaches, so, you know, the, the, the traditionals, or the traditional ones are X-ray crystallography and NMR spectroscopy, so these are sort of the, the, the two main techniques, uh, that, uh, give us the 3D coordinates. But nowadays, uh, there is huge breakthrough in, uh, cryo-electron microscopy. So the, the more advanced methods that allow us to, uh, you know, to get-... get into the, uh, you know, 3D, uh, shapes of much larger molecules, molecular complexes. Just to give you, uh, one of the common examples, uh, for this year, right? So, so, the, the first experimental structure of a SARS-CoV-2 protein was the cryo-EM structure of the S protein, so the spike protein. And so, um, it was solved very quickly, and the reason for that is the advancement of the, uh, of this technology is, is pretty spectacular.

Lex Fridman 6:53 How many domains does the, uh... Is it more than one domain, the spike-

Dmitry Korkin 6:57 Oh, yes. Oh, yes.

Lex Fridman 6:58 ... the spike protein?

Dmitry Korkin 6:58 So, I mean, so, so it's, it's a very complex structure and, and-

Lex Fridman 7:01 It's complex.

Dmitry Korkin 7:02 Uh, we, you know... On top of the complexity of a single protein, right? So this, this structure is actually, is a complex, is a trimer. So it needs to form a trimer in order to function properly.

Lex Fridman 7:17 What's a complex?

Dmitry Korkin 7:18 So, a complex is agglomeration of, uh, multiple proteins. And so, uh, we can have the same protein copied in multiple, uh, you know, uh, made up in multiple copies and forming something that we called, uh, a homo oligomer. Homo means the same, right? So, so in this case, so, uh, uh, spi- the spike protein is the, is an example of a ho- homo tetra-, uh, homo trimer, sorry. So-

Lex Fridman 7:46 It means three copies of a protein?

Dmitry Korkin 7:48 Three copies, exactly.

Lex Fridman 7:48 In order to... (laughs)

Dmitry Korkin 7:50 Exactly. We have the, these three chains, the, the three molecular chains, uh, coupled, uh, together and performing the, the function. That's what, when, when you look at this protein from, from the top, you see a perfect triangle.

Lex Fridman 8:03 Yeah.

Dmitry Korkin 8:04 So, uh, but other, uh, you know, so other complexes are made up of, um, you know, different proteins. Uh, some of them are completely different, some of them are similar. The, the hemoglobin molecule, right? So it's actually, it's a protein complex. It's made of four basic subunits. Two of them, um, are identical to each other, and two other are identical to each other, but they're also similar to each other, which sort of, uh, gives us some ideas about the evolution of this, uh, you know, uh, of this, uh, molecule. And, uh, perhaps one of the hypotheses is that, you know, uh, in the past, it was just a homo tetramer, right? So four identical comp-, uh, copies, and then it became, you know, uh, sort of, uh, modify-, uh, it, it became mutated, uh, over the time and, and became more specialized.

Lex Fridman 9:00 Can we, uh, linger on the spike protein for a little bit? Uh, is, is there something interesting or like beautiful you find about it?

Dmitry Korkin 9:07 I mean, first of all, it's an incredibly challenging protein. And so we, as a part of, uh, our sort of, um, research to understand the structural basis of this virus, to sort of decode, structure decode every single protein, uh, in its proteome, um, we tr- you know, we've been working on this spike, uh, protein. And, uh, one of the main challenges was that, um, the, uh, cryo-EM, uh, data allows us to, uh, reconstruct or to obtain the 3D coordinates of roughly two-thirds of the protein. The rest of the one-third, uh, of this protein, it's a part that, uh, is buried into the, into the membrane-

Lex Fridman 9:57 Yeah.

Dmitry Korkin 9:57 ... of the virus and, uh, of the, of, of the viral envelope. And, uh, it also has a lot of unstable structures around it.

Lex Fridman 10:07 So it's chemically interacting somehow with whatever the heck it's connecting to?

Dmitry Korkin 10:10 It's... Yeah. So, so it... People are still trying to understand. So, so the, the nature of and the, the role of this, uh, you know, uh, of this, uh, one-third. 'Cause the, the top part, uh, you know, the, the primary function is to get attached to the, you know, uh, ACE2 receptor-

Lex Fridman 10:28 Mm-hmm.

Dmitry Korkin 10:28 ... human receptor. There is also beautiful m- you know, me- um, mechanics of how this thing happens, right? So because there are three different co- uh, copies of this, uh, chains or, you know, there are three different domains, right? So we're talking about domains. So, so this is the receptor binding domains, RBDs, that gets untangled and get ready to, to, to atta- to get attached to, to the receptor. And now, they are not necessarily going in a sync mode. As a matter of fact-

Lex Fridman 11:05 Say synchronous?

Dmitry Korkin 11:06 So, yes. So, and this is, this is where, you know-

Lex Fridman 11:09 Yeah.

Dmitry Korkin 11:10 ... the, another level of complexity comes into play because, you know, right now, what we see is... We typically see just one of the arms going out-

Lex Fridman 11:21 Yeah.

Dmitry Korkin 11:21 ... and getting ready to, to atta- to be attached to the, uh, uh, to the ACE2 receptors. However, there was a recent mutation that, uh, uh, people studied in that, uh, spike protein. And, um, uh, a, um, very recently, a group from, um, UMass Medical School, uh, we happened to collaborate with group, so this is the group of, uh, Jeremy Luban and, uh, and a number of, uh, other faculty. Um, they, uh, actually, uh, solved the, um, the mutated structure of the spike and they showed that actually, because of these mutations, you have more than one arms opening up. And so now, so you, so the frequency of two arms going up increa- increased quite, you know, drastically.

Lex Fridman 12:18 Oh, interesting. Does that, does that change the dynamic somehow, of how it's gonna-

Dmitry Korkin 12:21 It, it potentially can change the dynamics of... Because now you have two possible opportunities to get attached to the ACE2 receptor.It's a very complex molecular process, mechanistic process. But the first step of this process is the attachment of this spike protein, of the spike trimer, to the human ACE2 receptor. So, this is a molecule that sits on the surface of the human cell.

Lex Fridman 12:51 Yeah, so-

Dmitry Korkin 12:52 And that's essentially what initiates the h- what triggers the whole process of, you know, encapsulation.

Lex Fridman 12:58 Of, uh, uh, if this was dating, this would be the first date. So, this is the, uh, (laughs)

Dmitry Korkin 13:03 In a way, yes.

Lex Fridman 13:04 (laughs) So, is it, is it possible to have the spike protein just like floating about on its own, or does it need that intractability with, uh, uh, with the membrane?

Dmitry Korkin 13:14 Yeah, so it needs to be attached, at least as far as I know. But, uh, you know, when you get this thing attached on the surface, right? There is also a lot of dynamics on h- where it, how it sits on the surface, right?

Lex Fridman 13:28 Yeah.

Dmitry Korkin 13:28 So for example, uh, there was a recent work in, uh, again, uh, where people used the cryo-electron microscopy to get the first glimpse of the overall structure. It's a very low res, but you still get some interesting details about the surface, about what is happening inside. Because we have literally no clue, until recent work, about how the, the capsid is organized. How-

Lex Fridman 13:54 What's a capsid?

Dmitry Korkin 13:55 So a capsid is essentially, it's, uh, the inner core of the viral particle where the, uh, there is a, the RNA of the virus, and it's pro- protected by another protein, N protein, uh, uh, that essentially acts as a shield. But, you know, now we are learning more and more. So, it's actually, it's not just this shield. It's u- it's potentially is used for the stability of the outer shell of the f- uh, of the virus. So, it's, it's pretty complicated.

Lex Fridman 14:27 And does that mean understanding all of this is really useful for trying to figure out like developing a vaccine or some kind of drug to attack any aspects of this, right?

Dmitry Korkin 14:36 So, I mean, there are many different imp- implications to that. You know, first of all, you know, it's, it's important to understand the virus itself, right? So, you know, in order to, uh, to understand how it acts, what is the overall mechanis- mechanistic process of this virus replication, of this virus proliferation to the cell, right? So, so that's one, uh, aspect. The o- the other aspect is, you know, designing new treatments, right? So one of the, uh, possible treatments is, um, you know, designing nanoparticles. And so, so nanoparticles that will resemble the viral shape, that would have the spike integrated, and essentially would act, uh, as a competitor to the real virus by blocking the ACE2 receptors and thus preventing the real virus entering the cell.

Lex Fridman 15:30 Mm-hmm.

Dmitry Korkin 15:31 Uh, now, there are also, uh, you know, there is a very interesting, uh, direction in, uh, looking at the, at the membrane, at the envelope portion of the protein, and attacking its, uh, M protein. So, so there are, uh, you know,

Dmitry Korkin 15:48 to give you a, you know, sort of a brief overview, there are four structural proteins that these are the proteins that made up a structure of the virus. So, spike, S protein, that acts as a trimer, so it needs three copies. E, envelope protein, that acts as a pentamer, so it needs five copies-

Lex Fridman 16:12 Mm-hmm.

Dmitry Korkin 16:12 ... to act properly. M, uh, is a, uh, is a membrane protein, and it me- it forms dimers. And actually, it forms beautiful lattice, and this is something that we've been studying, and we are seeing it in simulations. It, it actually forms a very nice grid or, you know, uh, you know, threads, uh, you know, uh, of, of different dimers attached next to each other in a very regular pattern.

Lex Fridman 16:35 Just a bunch of copies of each other, and they naturally, when you have a bunch of copies of each other, they form an interesting lattice.

Dmitry Korkin 16:40 Exactly.

Lex Fridman 16:40 Mm-hmm.

Dmitry Korkin 16:41 And, and, y- y- you know, y- if you think about it, right? So, so, so this complex, you know, the vi- the viral shape needs to be organized somehow, self-organized somehow, right? So it, it, you know, if it was a completely random process, you know, you probably wouldn't have the, the, the envelope shell of the ellipsoid shape in there. You would have something, you know, uh, pretty random, right? Uh, shape. So there is some, you know, regularity in how this, uh, you know, uh, h- how these, uh, M dimers get to a- attach to each other in a very specific directed way.

Lex Fridman 17:20 Is that understood at all?

Dmitry Korkin 17:22 Uh, it's not understood. We are now, uh, we- we've been working in the past six months since, you know, we- we met actually.

Lex Fridman 17:30 (laughs)

Dmitry Korkin 17:30 This is where, where we started working on, on trying to understand the overall, uh, structure of the envelope and the, the key components that made up this, uh, you know, uh, structure.

Lex Fridman 17:41 Wait, does the envelope also have the lattice structure or no?

Dmitry Korkin 17:43 So, so the envelope is essentially is the outer shell of the viral particle. The N, the nucleocapsid protein, is something that is inside.

Lex Fridman 17:54 Got it.

Dmitry Korkin 17:54 But get that. The N is likely to interact with M.

Lex Fridman 17:59 Does it go M and E? Like, where's the E and the M-

Dmitry Korkin 18:02 So, so E, those different proteins, they occur in different copies on the viral particle. So, so E, this pentamer complex, we only have two or three maybe per each particle.

Lex Fridman 18:19 Mm-hmm.

Dmitry Korkin 18:19 Okay? We have 1,000 or so of M dimers that essentially made up, uh, that makes up, uh, the entire, you know, outer shell.

Lex Fridman 18:30 For sure. So most of the outer shell is the M-

Dmitry Korkin 18:33 M dimer and lipids.

Lex Fridman 18:35 ... is the M protein. When you say particle, that's the virion, the virus, the individual virus-

Dmitry Korkin 18:40 It's a single, yes. Single...

Lex Fridman 18:41 ... single element of the vir-

Dmitry Korkin 18:42 Single... exactly.

Lex Fridman 18:43 Single virus.

Dmitry Korkin 18:43 Single virus.

Lex Fridman 18:44 Yeah.

Dmitry Korkin 18:44 Right. And we have about, you know, roughly 50 to 90 spike trimers.Right? So- so- so when you, you know, when you sh- show a c-

Lex Fridman 18:53 Per, per virus particle.

Dmitry Korkin 18:55 Per virus particle.

Lex Fridman 18:56 Fi- sorry, what did you say, 50 to 90?

Dmitry Korkin 18:58 50 to 90, right?

Lex Fridman 18:59 Cool.

Dmitry Korkin 19:00 So- so this is how this thing is organized. And so now, typically, right, so you see this, uh, uh, the, uh, the antibodies that target, you know, spike protein, certain parts of the spike protein, but there could be some, uh, also some treatments, right? So- so these are, uh, you know, these are small molecules that bind strategic parts of this proteins, disrupting its function. So one of the promising directions, um, it's one of the newest directions, is actually targeting the M-dimer of the protein, targeting the proteins that make up this outer shell, because if you're able to destroy the outer shell, you are essentially destroying the- the- the viral particle itself.

Lex Fridman 19:52 Mm-hmm.

Dmitry Korkin 19:52 So preventing it from- from, you know, functioning a- at all.

Lex Fridman 19:56 So that's, you think is, uh, from a sort of cyber security perspective, virus security perspective, that's the best attack vector, is, uh-

Dmitry Korkin 20:05 It's- it's- it's-

Lex Fridman 20:05 ... or like this, uh, that's a promising attack vector? It's-

Dmitry Korkin 20:08 I- I would say yeah. So I mean, it's just there's still tons of research needs to be s- you know, to be done. But, uh, yes. I think, you know, so-

Lex Fridman 20:16 There's more attack surface, I guess. (laughs)

Dmitry Korkin 20:19 More attack surface, but, you know, from- from our analysis, from other evolution analysis, this protein is evolutionary more stable compared to the, say, to the spike protein.

Lex Fridman 20:31 Uh, and stable means a- a more, uh, static target? (laughs)

Dmitry Korkin 20:35 Well, yeah, so- so it- it- it- it doesn't change. It doesn't evolve from the evolutionary s- uh, perspective so drastically as, for example, the spike

Dmitry Korkin 20:45 protein.

Lex Fridman 20:46 There's a bunch of stuff in the news about mutations of the virus in the United Kingdom.

Dmitry Korkin 20:51 Yep.

Lex Fridman 20:51 I also saw in South Africa something, maybe that was yesterday. Uh, you just kind of mentioned about stability and so on. Which aspects of this are mutatable, and which aspects, if mutated, become more dangerous? And maybe even zooming out, what are your thoughts and knowledge and ideas about the way it's mutated, all the news that we've been hearing? Are you worried about it from a biological perspective? Are you worried about it from a human perspective?

Dmitry Korkin 21:21 So I mean, you know, mutations are sort of a general way for these viruses to evolve, right? So- so it's, you know, it's, uh, essentially, this is the way they evolve. This is the way they were able to jump from, you know, one species to another. We also see, uh, you know, some recent jumps. There were some incidents, uh, o- of this virus jumping from human to dogs. So, you know, there i- there is some danger in- in- in- in those jumps, because, you know, every time it jumps, it also mutates, right? So- so it, when it jumps to- to the, uh, to the species and jumps back, right?

Lex Fridman 22:06 Yeah.

Dmitry Korkin 22:06 So it acquires some mutations that are sort of, um, driven by the environment of a new host.

Lex Fridman 22:15 Yeah.

Dmitry Korkin 22:16 Right? And it's different, uh, from the human environment. And so we don't know whether the mutations that are acquired, uh, in the new species are neutral with respect to the human host, or maybe, you know, maybe, um, damaging.

Lex Fridman 22:32 Yeah. Change is always scary. But-

Dmitry Korkin 22:34 Yep.

Lex Fridman 22:34 ... s- s- so are you worried about ... I mean, it seems like because the spread is during winter now, is seems to be exceptionally high, uh, and, uh, especially with a vaccine just around the corner, already being actually deployed, uh, there's some worry that the, there's, this puts evolutionary pressure, selective pressure on the virus e- in for it to, uh, to mute, uh, for it to mutate. Is that a s- source of worry?

Dmitry Korkin 23:00 Yes. Well, I mean, there is always this thought, uh, you know, in- in- in the scientist mi- mind, uh, you know, what happ- what will happen, right? So, uh, I know there've been, uh, there've been discussions about sort of the arms race between the, you know, the ability of- of the, uh, of the, uh, you know, humanity to, uh, you know, to get vaccinated faster than the virus, you know, uh-

Lex Fridman 23:28 (laughs)

Dmitry Korkin 23:28 ... essentially, uh, uh, you know, becomes, uh, you know, resistant to- to the vaccine. Um, I, I mean, I don't worry that much, uh, simply because, uh, you know, there is not that much evidence to that.

Lex Fridman 23:47 Too aggressive mutation around a vaccine?

Dmitry Korkin 23:50 Exactly. You know, obviously, there are mutations around the vac- uh, you know, uh, there are vaccines, so the reason we get vaccinated every year against the seasonal flu-

Lex Fridman 24:01 It's because there's mutations.

Dmitry Korkin 24:02 Right? Um, but, uh, you know, I- I think it's important to study it, no doubts, right? So- so I think one of the, you know, to me, and, uh, again, I might be biased, uh, because, you know, we- we- we- we- we've been, uh, trying to- to do that as well, uh, so but one of the critical directions in understanding the virus is to un- uh, to understand its evolution in order to, um, sort of understand the mechanisms, the key mechanisms that lead the virus to jump, you know, the Nordic viruses to jump from species, from species to another, that, uh, the mechanisms that lead the virus to become resistant, uh, to vaccines, also to treatments, right? And hopefully, that knowledge was, uh, will enable us to sort of forecast the evolutionary, uh, traces, the future evolutionary traces of this virus.

Lex Fridman 24:58 I mean, what, uh, from a biological perspective, this might be a dumb question, but is there parts of the virus that if, uh, souped up...... like, through mutation, could make it more effective at doing its job? We're talking about this specific coronavirus, like-

Dmitry Korkin 25:15 Yeah.

Lex Fridman 25:15 ... 'cause we were talking about the different, like the membrane, the M protein, the E protein, the N and, uh, S, uh, the spike. Is there some, uh-

Dmitry Korkin 25:26 And there are 20 or so more-

Lex Fridman 25:28 (laughs)

Dmitry Korkin 25:28 ... in addition to that.

Lex Fridman 25:30 But is that, is that a dumb way to look at it? Like, uh, which of these, if mutated, could have the greatest impact, potentially damaging impact, on the effectiveness of the virus?

Dmitry Korkin 25:43 So, i- i- it's actually, it's a very good question, uh, because... And, and the short answer is, we don't know yet. But, uh, of course there is capacity of this virus to, to become more efficient. The reason for that is, um, you know, so if you look at the virus, I mean, it's, it's a machine, right? So it's a machine that does a lot of different functions, and many of these functions are sort of nearly perfect, but they're not perfect. And, uh, those mutations can make those functions more perfect. For example, the attachment to ACE2 receptor, right? Of the spike. Right? So, uh, you know, is it... Has this virus reached the efficiency in which the attachment is carried out? Or there are some mutations that, uh, that still to be discovered, right, that will, uh, make this attachment, uh, sort of, uh, stronger or, you know, something, uh, more, in a way, more efficient from the point of view of this virus functioning. That's, that's sort of, uh, the obvious example, but if you look at each of these proteins, I mean, it's there for a reason, it performs certain function. And, uh, it could be that certain mutations will, you know, enhance this function. It could be that some mutations will make this function much less efficient-

Lex Fridman 27:13 Right.

Dmitry Korkin 27:13 ... right? So that's, that's also the case.

Lex Fridman 27:16 Let's, uh, s- since we're talking about the evolutionary history of a virus, uh, le- let's zoom back out and, uh, look at the evolution of proteins. I, I glanced at this, uh, 2010 Nature paper on the, quote, "Ongoing expansion of the protein universe."

Dmitry Korkin 27:34 Yeah.

Lex Fridman 27:34 And then, you know, it kind of implies and, uh, talks about that, uh, proteins started with a common ancestor, which is k- you know, kind of interesting. It's interesting to think about like, even just like, the first organic thing that started life on Earth, and from that, there's now, uh, you know, what is it? 3.5 billion years later, there's now millions of proteins, and they're still evolving. And that's, you know, in part one of the things that you're researching. Is there something interesting to you about the evolution of proteins from this initial ancestor, uh, to today? Is there something beautiful, insightful, about this long story?

Dmitry Korkin 28:18 So I think, you know, um, if, if I were to pick a single keyword about, uh, protein evolution, I would pick, uh, modularity, something that we talked about, uh, in the b- in the beginning, and that's the fact that the proteins are no longer considered as, you know, as a sequence of letters. There are hierarchical, uh, complexities in the way these proteins are organized, and, uh, these complexities are g- actually going beyond the protein sequence. It's actually going all the way back to the, uh, to the gene, to the nucleotide sequence. And so, you know, again, these protein domains, they are not only functional building blocks, they are also evolutionary building blocks. And so what we see in the, sort of in the later stages of evolution, I mean, once this stable s- structurally and functionally b- building blocks were discovered, they essentially, they s- stay, those domains stay as such. So, that's why if you start comparing, uh, different proteins, you will see that, uh, many of them will have similar fragments, and those fragments will correspond to something that we call protein domain families. And so, so they are still different, because you, you still have mutations and, and, and, and the, you know, uh, the, uh, you know, different mutations are attributed to, to, you know, diversification of the function of this, uh, you know, uh, protein domain. However, you don't... You very rarely see, um, you know, the, the evolutionary events that would split this domain into fragments, because... And it's, you know, uh, once you have the, the, the, the, uh, pro- the domain split, you actually, you, um, you know, you can completely cancel out its function, or at the very least, you can reduce it. And that's not, you know, efficient from the point of view of the, you know, of the cell functioning. So, so the, the, the protein domain level is a very important one. Now, on top of that, right? So if you look at the proteins, right? So you have these structural units, and they carry out the function. But then, uh, much less is known about things that connect these protein domains-

Lex Fridman 30:54 Mm.

Dmitry Korkin 30:54 ... something that we call linkers. And those linkers are completely flexible, you know, parts of the protein that nevertheless carry out a lot of function.

Lex Fridman 31:06 It's like little tails, little heads of pro-

Dmitry Korkin 31:08 So, so, so we, we do have tails, so they're called termini. C and N termini. So these are things right on the, on, on, uh, on, uh-... one and another ends of the protein sequence. So, they are also very important. So, they, they attributed to, to very specific, uh, interactions between the proteins. So-

Lex Fridman 31:28 But you're referring to the links between domains.

Dmitry Korkin 31:30 That connect the domains. And, you know, apart from the, just the- the, uh, simple perspective, if you have, you know, a very short domain, you have, uh, sorry, a very short linker, you have two domains next to each other. They are forced to be next to each other. If you have a very long one, you have the domains that are extremely flexible, and they carry out a lot of, sort of, spa- spatial reorganization, right?

Lex Fridman 31:56 That's awesome. (laughs)

Dmitry Korkin 31:58 So... But on top of that, right, just this linker itself, because it's so flexible, it actually can adapt to a lot of different shapes. And therefore, it's a, it's a very good interactor when it comes to interaction between this protein and other protein. All right? So these things also evolve, you know, uh, and they, in a way, have different lo- uh, sort of, uh, uh, laws of, uh, uh, or the driving laws that, uh, underlie the- the evolution, because they no longer need to, uh, to, uh, preserve certain structure, right? Uh, unlike protein domains. And so now, on top of that, you have, uh, something that is even less studied, and this is something that, um, uh, uh, attributed to- to the concept of alternative splicing.

Lex Fridman 32:52 Hmm.

Dmitry Korkin 32:53 So alternative splicing. So it's, uh, it's, uh, it's a very cool c- uh, concept. It's something that, uh, uh, we've been fascinated about for, you know, over a decade, uh, in my lab, and trying to do research with that. But, so, you know, so, so typically, you know, a- a v- a simplistic s- perspective is that one gene is equal one protein product. All right? So you have a gene, e- you know, you transcribe it and- and- and translate it, and you... It becomes a protein. In reality, when we talk about, uh, eukaryotes, especially sort of, uh, more recent eukaryotes that are very complex, the gene is no, is no longer equal to one protein. It actually can, uh, produce multiple functionally, uh, you know, active protein products, and each of them is, uh, you know, i- i- i- is, uh, uh, called an alternatively spliced product. The reason it happens is that if you look at the gene, it actually has... It has also blocks. And the blocks, some of which... And it- i- it essentially, it goes like this. So, we have a- a block that will later be translated. We call it exon. Then we'll have a block that is not translated, cut out.

Lex Fridman 34:19 Mm-hmm.

Dmitry Korkin 34:19 We call it intron. So we have exon, intron, exon, intron, et cetera, et cetera, et cetera, right?

Lex Fridman 34:24 Mm-hmm.

Dmitry Korkin 34:24 So, and some- sometimes you can have, uh, you know, dozens of these exons and introns.

Lex Fridman 34:29 Mm-hmm.

Dmitry Korkin 34:29 So what happens is, during the p- the process when the gene is converted to RNA, we have things that are cut out, the introns that c- uh, cut out, and exons that now get assembled together.

Lex Fridman 34:46 Mm-hmm.

Dmitry Korkin 34:47 And sometimes we will throw out some of the exons, and the remaining, uh, protein product will become-

Lex Fridman 34:54 Will still be the same.

Dmitry Korkin 34:55 ... diff- different.

Lex Fridman 34:56 Oh, different.

Dmitry Korkin 34:56 Right? So, so now you have, uh, fragments of the protein that no longer there. They were cut out with the introns.

Lex Fridman 35:03 Okay.

Dmitry Korkin 35:04 Sometimes you will s- uh, essentially take one exon and replace it with another one. Right?

Lex Fridman 35:09 So there's some flexibility in h- in this process.

Dmitry Korkin 35:12 So, so that creates a whole new level of complexity, 'cause now-

Lex Fridman 35:17 But this is random, though? Is it random?

Dmitry Korkin 35:19 Uh, uh, it's, it's not random. We... And, and this is where I think, uh, now the- the appearance of this modern, uh, single-cell, uh, uh, and- and, uh, before that, tissue, uh, level sequencing, next-generation sequencing techniques, such as RNA-Seq, allows us to see that these, these are the events that often happen in response, in... It's a, it's a dynamic event that happens in response to, to disease, or in response to certain developmental stage of a cell. And, and this is an incredibly complex layer that also undergoes... I mean, because it's, uh, at the gene level, right? So it undergoes certain evolution, right? And, uh, now we have this interplay between what happening, uh, what is happening in the, in the protein's world and what is happening in the, in the gene and, you know, RNA world. And for example, you know, it's- it's often that we see that the boundaries of these exons coincide with the boundaries of the protein domains.

Lex Fridman 36:30 Hmm.

Dmitry Korkin 36:31 Right? So, so there, there is, there is, uh, you know, uh, close interplay to that. Uh, it's not always. I mean, you know, uh, otherwise it would be too simple, right? But we do see the connection between those sort of machineries. And obviously the evolution will pick up this complexity and, uh, you know-

Lex Fridman 36:51 Select for whatever is successful, whatever

Narrator 36:53 Is interesting function.

Dmitry Korkin 36:53 Yeah. We will... We, we, we see that complexity in play, and- and makes this question, you know, more complex but more exciting.

Lex Fridman 37:02 A- a small detour. I don't know if you think about this. In- into the world of computer science, there's, uh, D- Douglas Hofstadter, I think, came up with a, uh, a name of Quine, which are... I don't know if you're familiar with these things, but it's computer programs that have, uh, I guess exon and intron, and they copy... The whole purpose of the program is to copy itself.

Dmitry Korkin 37:26 Mm-hmm.

Lex Fridman 37:26 So it prints copies of itself, but can also carry information inside of it. So it's a very kind of crude, fun exercise-

Dmitry Korkin 37:34 Oh, yeah.

Lex Fridman 37:35 ... of, um-... can we sort of replicate these ideas from cells, of can we have a computer program that when you run it, just prints itself?

Dmitry Korkin 37:44 Yes. Yes.

Lex Fridman 37:44 The who- the entirety of itself, and does it in different programming languages and so on? I've been playing around and writing them. It's a kind of fun little exercise.

Dmitry Korkin 37:53 You know, when I was a kid, so, so you know, it, it was essentially one of the, of the sort of main stages in, in, uh, informatics Olympiads-

Lex Fridman 38:06 Mm-hmm.

Dmitry Korkin 38:06 ... that you have to reach-

Lex Fridman 38:08 (laughs)

Dmitry Korkin 38:08 ... in order to be any so good-

Lex Fridman 38:10 Yeah.

Dmitry Korkin 38:11 ... is you should be able to write a program that replicates itself. And so, the task then becomes even, you know, sort of more complicated. So, what is the shortest-

Lex Fridman 38:22 What is the shortest, yeah.

Dmitry Korkin 38:24 ... program? And of course, it's, it's, you know, it's a function of a programming language. But yeah, I, I remember, you know, a long, long, long time ago, when we tried to, you know, to, to make it shorter and shorter-

Lex Fridman 38:34 (laughs)

Dmitry Korkin 38:34 ... and find the, the, the, the shortcuts.

Lex Fridman 38:36 There's actually, on, uh, Stack Exchange, there's a, an entire site called Codegolf, I think, where the entirety is just a competition. People just come up with the, whatever task, uh, that I don't know, like, uh, write code that reports the weather today. And then, the competition is about, in whatever programming language-

Dmitry Korkin 38:58 Shortest.

Lex Fridman 38:58 ... what is the shortest program, and it makes you actually ... People should check it out, because it makes you realize there's some, some weird (laughs) programming languages out there. Um, but you know, just to dig on that a little, uh, deeper, uh, do you think ... You know, in computer science, we don't often think about programs. There's like, the machine learning world now, uh, that's still kind of basic programs, and then there's humans that replicate themselves, right? And there's these mutations and so on. Do you think we'll ever have a world where there's programs that kinda have an evolutionary process? So, I'm not talking about evolutionary algorithms, but I'm talking about programs that kind of mate with each other and evolve, and like, on their own, replicate themselves? So, this is kind of ... Uh, the idea here is, you know, that's how you can have a runaway thing. So, we think about machine learning as a system that gets smarter and smarter and smarter and smarter. At least the machine learning systems of today ha- are like, it's, it's a program that you can, like, turn off, as opposed to throwing a bunch of little programs out there and letting them, like, multiply and mate and evolve and replicate. Do you ever think about that kind of world, you know, when we jump from the biological systems that, uh, you're looking at to, to artificial ones?

Dmitry Korkin 40:27 I mean, it's almost like you, you take the, the sort of the area of intelligent agents, right?

Lex Fridman 40:34 Yes.

Dmitry Korkin 40:34 Which are essentially the, the independent sort of, uh, codes that run and interact-

Lex Fridman 40:40 Yeah.

Dmitry Korkin 40:40 ... and exchange the information, right? So, I, I don't see why not. I mean, I, I, you know, it could be sort of a, a natural evolution in, in, in this, you know, uh, area of computer science.

Lex Fridman 40:53 I think it's kinda an interesting possibility. It's terrifying, too. But I think it's a really powerful tool, like to have like agents that inter- ... You know, we have social networks with millions of people, and they interact. I think it's interesting to inject into that what is already injected into that bots, right? But those bots are pretty dumb. Uh, uh, you know, they're, they're probably pretty dumb algorithms. Uh, y- you know, it's interesting to think that there might be bots that evolve together with humans, and there's this sea of humans and robots that are o- operating first in a digital space, and then you can also think ... I, I love the idea, some people who work, I think, uh, at Harvard, at Penn, there's, uh, robotics labs that, you know, build- ... Ta- it takes as a fundamental task to build a robot that, given extra resources, can build another copy of itself.

Dmitry Korkin 41:44 Mm-hmm.

Lex Fridman 41:45 Like, in the physical space.

Dmitry Korkin 41:46 Yep.

Lex Fridman 41:46 Which is, uh, super difficult to do, but super interesting.

Dmitry Korkin 41:50 Yes.

Lex Fridman 41:50 I remember there's like, research on l- robots that can build a bridge, so they make a copy of themselves-

Dmitry Korkin 41:57 Mm-hmm.

Lex Fridman 41:57 ... and they connect themselves.

Dmitry Korkin 41:58 Yeah.

Lex Fridman 41:58 And then, it sort of is like self-building bridge, based on building blocks. You can imagine like, a building that self-assembles.

Dmitry Korkin 42:05 Yes.

Lex Fridman 42:05 So it's basically self-assembling structures from, uh, from, uh, robotic parts. But it's interesting to, within that robot, add the ability to mutate, and, uh, and, (laughs) and do all the interesting like, little things th- that you're referring to in evolution-

Dmitry Korkin 42:23 Right.

Lex Fridman 42:23 ... to go from a single origin protein building block to like, this-

Dmitry Korkin 42:27 Well-

Lex Fridman 42:28 ... weird complexity.

Dmitry Korkin 42:28 And, and if you think about this, I mean, you know, the bits and pieces are there, you know, so, so you mentioned-

Lex Fridman 42:35 Yeah, right. (laughs)

Dmitry Korkin 42:35 ... the evolutionary algorithm, right? You know, so this is sort of, yeah ... And it, the, the, maybe it's sort of the, the, the goal is, in a way, different, right? So, the goal is to, you know, to essentially, uh, to, to optimize your search.

Lex Fridman 42:49 Right.

Dmitry Korkin 42:49 Right? So, uh, but, uh, sort of the, the ideas are there, so, so you, people recognize that, you know, that, uh, the re- rec- recombination events lead to global changes in the, in, in, in the s- search trajectories. The mutations event is a more refined, uh, you know, uh, step, uh, in the search. Then you have, you know, uh, other sort of, uh, nature-inspired algorithm, right? So, so one of the reason that, uh, that I, you know, I think it's, it's one of the funnest one is the slime, uh, based algorithm, right? So the-

Lex Fridman 43:25 (laughs)

Dmitry Korkin 43:25 ... the, the, it's, uh, you know, I think the- it first was introduced by the Japanese, uh, group, but where it was able to s- to solve, uh, some, some pretty, you know, complex problems. Uh, so, so that's the, uh ... And, and then I think, uh ...... there are still a lot of things we've yet to, to, you know, borrow from the nature, right? So, there are a lot of sort of ideas that nature, uh, you know, gets to offer us that, you know, it's up to us to grab it and to, to, you know, get the best use of it.

Lex Fridman 44:02 Including neural networks? You know, that we have a very crude inspir- inspiration from nature on neural networks. Maybe there's other inspirations to be discovered in the brain, or other aspects of, uh, the various systems, even like the immune system, the way it- it, uh, interplays. I recently, uh, started to understand that the- like, the immune system has something to do with the way the brain operates. Like, there's multiple things going on in there, which, uh, all of which are now modeled in artificial neural networks. And maybe if you throw a l- a little bit of that biological spice in there, you'll come up with something, uh,

Lex Fridman 44:38 something cool. I- I- uh, I'm not sure if you're familiar with the Drake equation, that, uh, estima- I just did a video on it yesterday because I wanted to give my own estimate of it. It's, uh, it's an equation that combines a bunch of factors to estimate how many alien civilizations are-

Dmitry Korkin 44:55 Oh, yeah.

Lex Fridman 44:56 ... in the galaxy.

Dmitry Korkin 44:56 I've- I've heard about it, yes, yes.

Lex Fridman 44:58 So one- one of the interesting parameters, you know, it's like, uh, how many, uh, stars are born every year, how many planets are on average per star, uh, for the s- how many habitable planets are there.

Dmitry Korkin 45:13 Mm-hmm.

Lex Fridman 45:14 And then, the- the one that starts being really interesting, uh, is, uh, the probability that life emerges on a habitable planet. So like, I don't know if you think about... You certainly think a lot about evolution, but do you think about the thing which evolution doesn't describe, which is like, the beginning (laughs) of evolution, the origin of life. I think I put the probability of life developing in a habitable planet at 1%. This is very scientifically, uh, rigorous. Okay, uh, what... First, at a high level for the Drake equation, what would you put that percent at on Earth? And in general, do you have something... Do you have thoughts about how life might have started? You know, like the proteins being the first kind of, one of the early jumping points?

Dmitry Korkin 46:02 Yeah, so- so, um, I think back in 2018, there was a very exciting paper published in Nature, where they, um, found, uh, one of the simplest amino acids, glycine, in this- i- in a comet dust.

Lex Fridman 46:23 Mm-hmm.

Dmitry Korkin 46:23 So- so this is, uh, and I apologize if I, uh, don't pronounce... It's a Russian named comet, it's I think Churyumov-Gerasimenko. This is the comet where, uh, and there was this, uh, um, mission to- to get, uh, and, uh, get close to this comet and get the- the, uh, stardust from- from its tail. And, uh, when scientists analyzed it, they actually found traces of, uh, you know, uh, of glycine, which, you know, makes up, you know, the on- it's one of the basic, uh, one of the 20 basic, uh, amino acids that makes up proteins, right? So, uh, so that was kind of-

Lex Fridman 47:08 That's exciting.

Dmitry Korkin 47:09 ... very exciting, right? But, uh, you know, it's- uh, the question is very interesting, right? So what, uh, you know, what... If there is some alien life, w- is it gonna be made of proteins?

Lex Fridman 47:22 Yeah.

Dmitry Korkin 47:22 Right? Or maybe RNAs, right? So we see that, you know, the- the RNA viruses are certainly, you know, very well-established sort of, um, you know, group of molecular machines.

Lex Fridman 47:36 Mm-hmm.

Dmitry Korkin 47:37 Right? So, um, so yeah, it's- it's- it's- it's a very interesting question, you know, yeah.

Lex Fridman 47:42 What- what probability would you put? Like how hard is this j- like, uh, how unlikely just on Earth do you think this whole thing is that we got going? Like is- are we really lucky or is it inevitable? Like what's your sense when you sit back and think about life on Earth? Is it higher or lower than 1%? Well, because 1% is pretty low, but it still is like, damn, that's pretty good chance.

Dmitry Korkin 48:05 Yes, it's- it's a pretty good chance.

Lex Fridman 48:06 (laughs)

Dmitry Korkin 48:06 I mean, I w- I- I would personally... But again, you know, uh, (laughs) I'm, um, you know, probably not the best person to- to- to- to do such estimations, but, uh, I would, you know... Intuitively, I would probably put it lower.

Lex Fridman 48:22 Lower.

Dmitry Korkin 48:23 But still, I mean, you know, given-

Lex Fridman 48:24 So we're really lucky here on Earth.

Dmitry Korkin 48:27 Uh, I mean...

Lex Fridman 48:28 Or the conditions are really good. I mean, that's...

Dmitry Korkin 48:31 It's this, you know, I think that there was- everything was right in a way, right? So- so we still, it's not the- the conditions were not like ideal if you try to- to look at, you know, what was, you know, several billions years ago when- when the life merged.

Lex Fridman 48:48 So there- there is something called, uh, the Rare Earth Hypothesis that, you know, encounter to the Drake equation says that the, you know, the conditions of Earth, if you actually were to describe Earth, it's quite a special place, so special it might be unique in our galaxy and potentially, you know, close to unique in the entire universe. Like it's very difficult to reconstruct those same conditions. And what the Rare Earth Hypothesis argues is all those different conditions are essential for life. And so that's the- it's sort of the counter, you know, like all the things we... Thinking that Earth is pretty average, um, I mean, I can't really... I'm trying to remember to- to go through all of them, but just the fact that it, um, is shielded from a lot of asteroids, the- obviously the distance to the sun, but also the fact that it, um...... it's like a perfect balance between the amount of water and land, and all those kinds of things. I d- I don't know, there's a bunch of different factors that, uh, I d- I don't remember. There's a long list. But it's fascinating to think about if, uh, if e- if, uh, in order for something like proteins and then DNA and RNA to, to emerge, you need, um, and basic living organisms, you need to be a very close and Earth-like planet, which would be sad.

Dmitry Korkin 50:18 Well, yeah.

Lex Fridman 50:18 Or, or exciting. I don't know which, uh ...

Dmitry Korkin 50:19 If you ask me, I, you know, in a way, I put the parallel between, um, you know, between our own research, uh, and I mean, from the, from the intuitive perspective, you know, you have those t- two extremes. And the reality is never ... very rarely-

Lex Fridman 50:40 Yeah. (laughs)

Dmitry Korkin 50:40 ... falls into the extremes. It's always the optimums always reached somewhere in between. So, so I would ... so, and that's what I tend to think. I think that, uh, you know, we're probably somewhere in between, so that we're not unique, unique. But again, the chances are, you know, reasonably small.

Lex Fridman 51:02 The problem is we don't know the, the other extreme, is like, I tend to think th- that we don't actually understand the basic mechanisms of like, what this has all originated from. Like, it seems like we think of life as this distinct thing. Maybe intelligence as a distinct thing. Maybe the physics that from which planets and suns are born is a distinct thing. But that could be a very ... it's like the Stephen Wolfram thing. It's like the, from simple rules emerges greater and greater complexity. So, uh, you know, I tend to believe that just life finds a way. It, like, we don't know the extreme of how common life is, 'cause it could be life is like, everywhere. (laughs) Like, like, so everywhere that it's almost like laughable. Like, that we're such idiots to think we're u- like, it's, it's, it's like ridiculous to even, like, think. It's like ants thinking th- that their little colony is the unique thing and everything else doesn't exist. I mean, it, it's also very possible that that's, uh, that's the extreme, and we're just not able to maybe comprehend the nature of that, uh,

Lex Fridman 52:11 life. I ... uh, just to stick on alien life for just a brief moment more, is, there is some signs of, uh, signs of life on Venus in gaseous form. There's s- uh, hope for life on Mars, probably extinct. We're not talking about intelligent life. (laughs) Although that has been in the news recently. (laughs) We're talking about basic like, you know, uh, bacteria kind of thing.

Dmitry Korkin 52:37 Little bacteria.

Lex Fridman 52:37 Yeah. Uh, and then also, I guess, uh, uh, there's a couple moons that I guess-

Dmitry Korkin 52:42 Europa.

Lex Fridman 52:42 Yeah, Eu- Europa, which is Jupiter's moon.

Dmitry Korkin 52:46 Yep.

Lex Fridman 52:46 I think there's another one. Are you, um ... is that exciting or is it terrifying to you that we might find life? Do you hope we find life?

Dmitry Korkin 52:54 I certainly do hope that we find life. Um, I mean, it was very exciting to, to hear about, uh, you know, uh, this, uh, news about the, the l- possible life on Venus.

Lex Fridman 53:09 It'd be nice to have hard evidence of something with, uh ... which is what the hope is for, for Mars and, and, uh, Europa. But do you think those organisms would be similar biologically, or would they even be sort of carbon-based if we do find them?

Dmitry Korkin 53:25 I would say they, they would be carbon-based. Uh, how similar, it's a big question, right? So, it's, it's, you know, the moment we discover things outside Earth, right, even if it's a tiny little single cell, I mean, there is so much ...

Lex Fridman 53:45 Just imagine that. That would be so ... (laughs)

Dmitry Korkin 53:47 It's ... uh, I, I think that, that would be another turning point for-

Lex Fridman 53:50 Yeah.

Dmitry Korkin 53:50 ... for the science, you know.

Lex Fridman 53:52 And if, especially if it's different in some very new way, that's exciting, 'cause that says ... that's a definitive sta- not a definitive, but a pretty strong statement that life is everywhere in the, in the, in the universe. To me at least, that's, that's really exciting.

Lex Fridman 54:09 You brought up Joshua Lederberg in an offline conversation. I think I'd love to talk to you about AlphaFold, and this might be an interesting way to enter that conversation because ... uh, so he won the 1958 Nobel Prize in physiology and medicine for discovering that bacteria can mate and exchange genes. But, uh, he also did a ton of other stuff, like, uh, like we mentioned, uh, helping NASA find life on Mars and, uh, the, uh ...

Dmitry Korkin 54:41 Dendral.

Lex Fridman 54:41 Dendral, the, the, the chemical expert system. Expert systems, remember those? Uh, do you, uh ... what do you find interesting about this guy and his, his ideas about artificial intelligence in general?

Dmitry Korkin 54:55 So I have a kind of personal story to, um, to share. So, I started my PhD in, in Canada back in 2000. And so essentially, my, my PhD was, uh, so we were developing sort of a, a new language for symbolic, uh, machine learning. So, it's, uh, different from the feature-based machine learning. And, and the, uh, one of the sort of, uh, cleanest applications of this, uh, you know, o- of this approach, of this formalism was, uh, to, uh, chem informatics and computer-aided drug design. Right? So, so, so essentially we were, uh, you know, so as a part of my research, um, uh, I developed a system that essentially looked at, uh, chemical compounds of, say, the same therapeutic category, you know, um, male hormones, right, and tried to, uh, figure out, um ...... the structural fragments that are the structural building blocks that are important that define this class, versus structural building blocks that are there just because, you know, the, to complete the structure. But they are not essentially the ones that make up the, the chemical, the, the key chemical properties of this, uh, therapeutic category. And, and, uh, you know, uh, for me, it was something new. I was, I was trained as a, an applied mathematician, you know, as with some machine learning background, but, you know, computer-aided drug design was completely, a completely new territory. So because of that, I often, uh, find myself asking lots of questions, uh, on one of the sort of central, uh, forums. Back then, there were no, no Facebooks or stuff like that. There was a forum, comp-

Lex Fridman 56:42 What's a forum? (laughs)

Dmitry Korkin 56:43 ... you know, it's, a forum is, it's essentially, it's like a bulletin board-

Lex Fridman 56:46 Yeah.

Dmitry Korkin 56:46 ... right, when, when you-

Lex Fridman 56:47 On the internet.

Dmitry Korkin 56:48 Yeah, so there, you essentially, you have a bunch of people and you post a question and you get, you know, an answer from, you know, different people. And, and, and back then, the s- one of the most popular, uh, forums was CCL, think, um, computational chemistry s- lib- uh, not library, but, uh, something like that. But CCL, that, that was the, the, the forum. And there I, I, you know, I w- was-

Lex Fridman 57:12 Asked a lot of dumb questions.

Dmitry Korkin 57:14 Yes, I asked questions, also shared some, some, you know, uh, some, uh, information about our formulas and how we do, and whether what- whatever we do makes sense. And so, you know, and, uh, I remember that, uh, one of these posts, I mean, I still remember, w- you know, uh, I, uh, I would call it desperately looking for, uh, for, uh, a chemist advice, something like that, right? And so, so I post my, uh, question, I explained ho- you know, how, how my f- uh, our formalism is, what is, what it does, and wha- what kind of applications I'm planning to, to do. And, you know, and it was, you know, in the middle of the night, and, you know, I went back, uh, you know, to bed. And n- and next morning, have a phone call from my advisor, who also looked at this forum. He's like, "You won't believe, uh, who, who replied to you." And, and it's like, "Who?" And he said, "Well, you know, there is a message, uh, to you from Joshua Lederberg." And my reaction was like, "Who is Joshua Lederberg?" (laughs) And...

Lex Fridman 58:24 Your advisor hung up. (laughs)

Dmitry Korkin 58:27 (laughs) So, and essentially, Jo- you know, Joshua wrote me that we, we had conceptually similar ideas in, in the Dendral Pro- uh, Project, you may wanna look it up.

Lex Fridman 58:38 Mm.

Dmitry Korkin 58:39 And, you know, and I-

Lex Fridman 58:40 And we, we should also, sorry, and as a side comment, say that even though he, he won the Nobel Prize at a really young age, in f- in '58, but so-

Dmitry Korkin 58:49 He, he was, I think, he was, what, 33.

Lex Fridman 58:52 Yeah, which is crazy.

Dmitry Korkin 58:53 Yep.

Lex Fridman 58:54 So anyway, so that's, so hence, hence in the '90s, (laughs) responding to young whipper-snappers on the, on the CCL forum. Okay, and so...

Dmitry Korkin 59:02 And, and, and so, so, so back then, he was already very senior. I mean, he unfortunately passed away back in 2008. Uh, but, you know, uh, back in 2001, he was, I mean, he, he was a professor emeritus at, uh, Rockefeller University. And, you know, that was actually, believe it or not, one of the, one of the, uh, of, uh, of the reasons I decided to join, uh, you know, as a postdoc, uh, the, the group of Andrei Sallou, who was at Rockefeller University-

Lex Fridman 59:30 Mm-hmm.

Dmitry Korkin 59:30 ... with the hope that, you know, that I could actually, you know, uh, have a chance, uh, to meet Joshua in person. And I met him very briefly, right? The, uh, just because he was walking, you know, uh, th- there's a little bridge that connects the, sort of the research ca- uh, campus with the, um, w- with the, uh, sort of skyscraper that Rockefeller o- owns, the, where, you know, uh, postdocs and faculty and, and graduate students live. And so, so I met him, you know, and had a very short, uh, conversation, you know. But, uh,

Dmitry Korkin 1:00:07 so I, I started, you know, reading about Dendral, and I was amazed. You know, it's, we're talking about 1960-

Lex Fridman 1:00:14 Yeah.

Dmitry Korkin 1:00:14 ... right? The ideas were so profound. Mm-hmm.

Lex Fridman 1:00:19 Well, what's the fundamental ideas of it?

Dmitry Korkin 1:00:21 The, the reason to make this is even crazier.

Lex Fridman 1:00:24 (laughs)

Dmitry Korkin 1:00:25 So, so, so Lederberg wanted to make a system that would help him study the, um, extraterrestrial molecules, right? So, so the idea was that, you know, the, the way you study the extraterrestrial molecules is you do the mass spec analysis.

Lex Fridman 1:00:45 Mm-hmm.

Dmitry Korkin 1:00:46 Right? And so the mass spec gives you sort of bits, numbers about essentially, um, gives you the ideas about the possible fragments or, you know, atoms and, you know, and, and, and, and maybe little fragments, pieces of this molecule that make up the molecule, right? So now you need to sort of, to, uh, decompose this information and to figure out what was the whole-

Lex Fridman 1:01:12 Mm-hmm.

Dmitry Korkin 1:01:12 ... before, you, you know, it beca- it became, uh, fragments, bits and pieces, right? So, so in order to make this, uh, you know, to have this tool, uh, the idea of Lederberg was to connect chemistry, computer science, and to design this so-called expert system that looks, that takes into account this, uh, takes as an input the mass spec data, the possible database of possible molecules, and essentially try to, um, sort of induce the molecule that would correspond to this spectra-

Lex Fridman 1:01:55 Mm-hmm.

Dmitry Korkin 1:01:56 ... or, you know, essentially...... the, what this project ended up being, uh, uh, was that, you know, it would provide a list of candidates that then a chemist would look at and, and, and make final decision. So-

Lex Fridman 1:02:12 But the original idea, I suppose, is to solve the entirety of this problem automatically?

Dmitry Korkin 1:02:16 Yes, yes. So, so, so he, uh, you know, so, uh, so he, uh, back then, uh, he-

Lex Fridman 1:02:24 60s. (laughs)

Dmitry Korkin 1:02:24 ... approached... Yes, believe that.

Lex Fridman 1:02:27 Yeah.

Dmitry Korkin 1:02:27 You know, it's, it's amazing. I mean, it still b-blows my mind, you know, that it's, that's, uh, it's, uh, an- and this was essentially the, the, the origin of the modern bioinformatics, cheminformatics, you know, back in '60s.

Lex Fridman 1:02:42 Yeah.

Dmitry Korkin 1:02:42 Right? So that's, that's, you know, uh, uh... So, every time you, you, you, you deal with, with projects like this, with the, you know, research like this, you just, you know, uh, so the, the power of, of the, of the, you know, intelligence of these people, uh, is, is just, you know, overwhelming.

Lex Fridman 1:03:01 Do you think about expert systems? Is there, um... And why they kind of didn't become successful? Especially in the space of bioinformatics, where it does seem like there is a lot of expertise in humans. And, uh, you know, it's f- it's possible to see that a system like this could be made very useful-

Dmitry Korkin 1:03:23 Right.

Lex Fridman 1:03:23 ... and be built up.

Dmitry Korkin 1:03:24 So, it's, it's actually, it's a, it's a great question. And, and this is something so... You know, so, uh, you know, uh, at my university, I teach ar- artificial intelligence. And, you know, we start the... My, my first two lectures are on the history of AI. And there, we, you know, we try to, uh, you know, go through the main stages of AI. And so, you know, the question of why expert systems failed or became obsolete, it's actually a very interesting one. And there are... You know, if you, uh, try to read the, you know, the historical perspectives, there are actually two lines of thoughts. One is that the- they were, uh, essentially not up to the expectations, and so therefore, they were replaced, you know, uh, by, by other things, right? The other one was that, uh, completely opposite one, that they were too good. And, a- and as a result, they essentially became sort of a household name, and then essentially, they, they got transformed. I mean, they, they... In both cases, sort of the outcome was the same, they evolved into something-

Lex Fridman 1:04:42 Yeah.

Dmitry Korkin 1:04:43 ... right? And that's what I... You know, if-

Lex Fridman 1:04:46 That's interesting.

Dmitry Korkin 1:04:46 ... if I look at this, right, so the modern machine learning, right?

Lex Fridman 1:04:50 Yeah.

Dmitry Korkin 1:04:50 So it-

Lex Fridman 1:04:51 So there's echoes in, in the modern machine learning to that.

Dmitry Korkin 1:04:53 I think so. I think so, because, you know, if y- if you think about this, you know, and how we design, uh, you know, uh, the most successful algorithms, including AlphaFold, right? You built in the knowledge about the domain, uh, that you study, right? So, so you built in your expertise.

Lex Fridman 1:05:12 So speaking of AlphaFold, so DeepMind's AlphaFold 2 recently, uh, was announced to have, quote-unquote, "solved protein folding." Um, how exciting is this to you? It s- seems to be one of the, one of the exciting things that have happened in 2020. It's an incredible accomplishment from the looks of it. What part of it is amazing to you? What part would you say is over-hype or maybe, uh, misunderstood?

Dmitry Korkin 1:05:38 Mm-hmm. It's definitely a very exciting achievement. To give you a little bit of perspective, right? So, uh, so in bioinformatics, we have, um, several competitions. And so the way, you know, you often hear, um, how those competitions have been explained to, uh, sort of to non-bioinformaticians is they, you know, they call it Bioinformatics Olympic Games. And there are several disciplines, right? So, so the, uh, so the, the... Historically, one of the first one was the discipline in predicting the protein structure.

Lex Fridman 1:06:09 Mm-hmm.

Dmitry Korkin 1:06:10 Predicting the 3D coordinates of the protein. But there are some others. So, uh, the predicting protein functions, uh, predicting effects of, uh, mutations on protein functions. Then, uh, predicting, uh, protein-protein interactions.

Lex Fridman 1:06:24 Mm-hmm.

Dmitry Korkin 1:06:25 So, so the original one was, uh, CASP, or, uh, Critical Assessment Of, uh, of, uh, Protein Structure. Um, and, um, the, you know... Typically, what, uh, happens during these competitions is, uh, you know, scientists, experimental scientists, uh, solve the- the structures, but don't put them into the protein data bank, which is a centralized database k- that contains all the 3D coordinates. Instead they hold it and, uh, release protein sequences. And now the challenge of the community is to predict-

Lex Fridman 1:07:07 Mm-hmm.

Dmitry Korkin 1:07:07 ... the 3D structures of these proteins, and then use the experimentally solved structures to assess which one is the closest one.

Lex Fridman 1:07:16 Right.

Dmitry Korkin 1:07:16 Right?

Lex Fridman 1:07:16 And this competition, by the way, just a bunch of different tangents, and maybe you can also say what is protein folding, uh, in this competition, CASP competition, is, has become the gold standard, and that's what was used to say that protein folding was solved. Sorry, just to add a little-

Dmitry Korkin 1:07:34 Yeah.

Lex Fridman 1:07:34 ... um, just a bunch... So if, if you could, whenever you say stuff, uh, maybe throw in some of the basics for the folks that might be outside-

Dmitry Korkin 1:07:40 Absolutely.

Lex Fridman 1:07:40 ... of the field. Anyway, sorry. For-

Dmitry Korkin 1:07:42 So, so yeah. So, you know, so the reason it's, it's, um, you know, it's relevant to our understanding of protein folding is because, you know, we, we, we've yet to learn how the folding mechanistically works, right? So there are different hypotheses.... w-what, what happens to this fold. For example, uh, there, there is a hypothesis that the folding happens by, you know, in-in... also in the modular fashion, right?

Lex Fridman 1:08:12 Mm-hmm.

Dmitry Korkin 1:08:12 So that, uh, we, we ha- we have protein domains that get folded independently, because, uh, their structure is stable and then the whole protein structure, uh, gets formed. But, you know, e- within those domains, we also have s- so-called secondary structure, the small alpha helixs, beta sheets. So these are, you know, uh, uh, elements that are structurally stable. And so, uh, and the, the question is, you know, when they... when do they get formed? Because some of the secondary structure elements, you have to have, uh, you know, a fragment in the beginning and, say, fra- uh, the fragment in the middle, right? So, so you cannot potentially start having, uh, the, the full fold from the get-go, right?

Lex Fridman 1:08:56 Mm-hmm.

Dmitry Korkin 1:08:56 Uh, so, so it's still, you know... I- it's still a big enigma, what, what happens. We know that it, it's an extremely efficient and stable process, right?

Lex Fridman 1:09:05 So there's, there's this long sequence, and the fold happens really quickly.

Dmitry Korkin 1:09:09 Exactly. So, yeah-

Lex Fridman 1:09:10 So that's really weird, right?

Dmitry Korkin 1:09:11 A- a- and-

Lex Fridman 1:09:11 And it happens, like, the same way almost every time.

Dmitry Korkin 1:09:15 Ex- exactly. Exactly. Right. So, so-

Lex Fridman 1:09:16 That's really weird. How-

Dmitry Korkin 1:09:17 So, so-

Lex Fridman 1:09:17 That's freaking weird. (laughs)

Dmitry Korkin 1:09:19 It's, it's... yeah. That's, that's why it's... it's, uh, it's such an am- amazing thing.

Lex Fridman 1:09:22 (laughs) Doesn't make any sense to me.

Dmitry Korkin 1:09:22 But most importantly, right? So, it's, you know... so when, when you see the, the, the, the tr- you know, the translation process, right? So when you don't have the, the, the whole, uh, protein translated, right? It's still being translated, you know, uh, uh, getting out from the r- ribosome, y- you already see some structural, you know, fragmentation. So, so folding starts happening before the whole protein gets produced, right?

Lex Fridman 1:09:53 Mm-hmm.

Dmitry Korkin 1:09:53 And so this is... this is obviously, you know, one of the biggest questions in, you know, in modern molecular biology.

Lex Fridman 1:10:01 Not, not like maybe what happens? Like, that's not... that's bigger than the question of folding. That's the question of, like... so like deeper fundamental idea of folding-

Dmitry Korkin 1:10:12 Yes.

Lex Fridman 1:10:12 ... behind folding.

Dmitry Korkin 1:10:13 Exactly, exactly. So, you know, so obviously, if we are able to, um, predict the end product-

Lex Fridman 1:10:22 Yeah.

Dmitry Korkin 1:10:22 ... of protein folding, we are one step closer to understanding sort of the mechanistics of the protein folding. Because we can then potentially look and, and start probing what are the critical parts of this process and what are not so critical pro- part of this pro- ... so now we s- we can start decomposing this-

Lex Fridman 1:10:44 Okay.

Dmitry Korkin 1:10:44 ... you know. So, so, so in a way, this protein structure prediction algorithm can be... can be used as a tool, right?

Lex Fridman 1:10:53 Mm-hmm.

Dmitry Korkin 1:10:53 So, so you change the, the, the... you know, you, you modify the, the protein, you get back to, to this tool. It predicts, okay, it's completely... it's completely unstable.

Lex Fridman 1:11:05 Yeah. Which, uh, which aspects of the input will have a big impact on the output.

Dmitry Korkin 1:11:09 Exactly, exactly. Exactly. So, so what happens is, you know, we typically have some sort of incremental, uh, advancement. You know, each stage of this, uh, CASP competition, you have groups with incremental advancement. And, you know, historically, uh, the top performing groups were, uh, you know... they were not using machine learning. They were using, uh, very advanced biophysics combined with bioinformatics, combined with, you know, the, the data mining, uh, and, uh, that was, uh, you know... that would enable them to obtain, uh, protein structures of those proteins that don't have any structurally solved relatives. Because, you know, if we have s- uh, a-another protein, say, the same protein, but coming from a different species, uh, we could potentially derive some ideas, and that's so-called homology or comparative modeling, where we'll derive some ideas from the previously known structures, and that would help us tremendously in, uh... you know, in, uh, reconstructing the 3D structure, uh, overall. But what happens when we don't have these relatives? This is when it becomes really, really hard, right? So that's so-called De Novo, uh, uh, you know, uh, De Novo protein structure prediction. And in this case, those methods were, uh, traditionally very good. But what happened in the... in the last year, the original AlphaFold came into, and all of a sudden, it's much better than e- everyone else.

Lex Fridman 1:12:56 This, this is 2018?

Dmitry Korkin 1:12:58 Yeah.

Lex Fridman 1:12:58 Oh, then... and the competition is only every two years-

Dmitry Korkin 1:13:00 Um-

Lex Fridman 1:13:00 ... I think.

Dmitry Korkin 1:13:02 And, and then... so, uh, you know, it was sort of... kind of of a shockwave to, to, to, to the bioinformatics community that, you know, we have, like, a state-of-the-art machine learning system that does, uh, you know, structure prediction. And, and essentially what it does, uh... you know, so, so, so, you know, if you look at this, it actually predicts the context. So, you know... so, so the, the process of, uh, reconstructing the, the 3D structure starts by predicting the, the context between the different parts of the protein-

Lex Fridman 1:13:38 Mm-hmm.

Dmitry Korkin 1:13:38 ... and the context is essentially the part of the proteins that are in the close proximity to each other.

Lex Fridman 1:13:43 Right. So it-

Dmitry Korkin 1:13:43 Right?

Lex Fridman 1:13:44 Uh, actually, the machine learning part seems to be estimating, you can correct me, uh, if I'm, I'm wrong here, but it seems to be estimating the distance matrix, which is like the distance between the different parts.

Dmitry Korkin 1:13:55 Yeah. So, so we call it the contact map.

Lex Fridman 1:13:58 Contact map.

Dmitry Korkin 1:13:58 Right? So once you have the contact map, the reconstruction is becoming more straightforward-

Lex Fridman 1:14:04 Yeah.

Dmitry Korkin 1:14:04 ... right? But so the contact map is the key. And so, so, uh, you know, so that...... what happened, and, uh, now we started seeing in this current stage, right, where in the, in the most recent one, we started seeing the emergence of these ideas in others, uh, people works.

Lex Fridman 1:14:24 Mm-hmm.

Dmitry Korkin 1:14:24 Right? But yet, here is, you know, AlphaFold2-

Lex Fridman 1:14:29 Yeah.

Dmitry Korkin 1:14:29 ... that again outperforms everyone else, you know, so by introducing yet another wave of, of, of the, of the, you know, machine learning ideas.

Lex Fridman 1:14:38 Yeah. Uh, there does seem to be also an incorporation, first of all, this, the paper is not out yet, but there's a bunch of ideas already out. There does seem to be an in- incorporation of this other thing, I don't know if it's something that you could speak to, which is like, the incorporation of like, other str- structures, like evolutionary similar...

Dmitry Korkin 1:15:00 Yes.

Lex Fridman 1:15:00 ... structures that are used to kind of give you hints.

Dmitry Korkin 1:15:03 Yes. So, so, so it's a, evolutionary similarity, uh, is something that we can detect at different levels, right? So, uh, we know for example that, uh, the structure of proteins is more conserved than the sequence. The sequence could be very different, but the structural shape is actually still very conserved. So that's, that's sort of the intrinsic property that, you know, in a way related to protein folds, i- you know, to the evolution of the, you know, of the protein, uh, of proteins and protein domains, et cetera. But we know that, I mean, we, there've been multiple studies. And, uh, you know, ideally if you have structures, you know, you should use that information. However, sometimes we don't have this information, instead we have a bunch of sequences, sequences we have a lot, right? So, so we, we, we have, you know, hundreds, thousands of, uh, you know, different organisms sequenced, right? And by taking the same protein, but in different organisms and aligning it, so making it, you know, making the corresponding positions aligned, we can actually, uh, say a lot about sort of what is conserved in this protein, and therefore, you know, structurally more stable, what is diverse in these proteins. So on top of that, we, we could provide sort of the information about the, sort of the secondary structure of this protein, et cetera, et cetera. So this information is extremely useful, and it's already there. So, so while it's tempting to, you know, to do a complete ab initio, so you just have a protein sequence and nothing else, the reality is such that we, we are overwhelmed with this data. So why not use it?

Lex Fridman 1:16:55 Mm-hmm.

Dmitry Korkin 1:16:56 And so yeah, so I'm looking forward to, to reading the, the, this paper.

Lex Fridman 1:17:01 It, it does seem to, like they've, in the previous version of AlphaFold, they didn't, uh, for this, the evolutionary similarity thing, they didn't use machine learning for that. Or they, rather they used it as like the input to the entirety of the, the neural net, like the features, uh, derived from the similarity. It seems like there is some kind of, quote unquote, iterative thing, where it seems to be, um, part of the, part of the learning process is the incorporation of this evolutionary similarity.

Dmitry Korkin 1:17:34 Yeah, I, I don't think there is a bio-archive paper, right?

Lex Fridman 1:17:37 There's nothing.

Dmitry Korkin 1:17:38 No. Nothing. Yeah, so-

Lex Fridman 1:17:38 There's a blog post that's written by a marketing team essentially.

Dmitry Korkin 1:17:42 Yeah.

Lex Fridman 1:17:42 Which, you know, it has some scientific, uh, uh, similarity probably to the, the actual methodology used, but it could be, it's like interpreting scripture. It could-

Dmitry Korkin 1:17:55 Yeah.

Lex Fridman 1:17:55 ... (laughs) it could be just poetic, uh, interpretations of the actual work, as opposed to direct connection to the work.

Dmitry Korkin 1:18:01 Yeah. So now, speaking about protein folding, right? So, so, so, you know, in order to answer the question whether or not we, we have solved this, right?

Lex Fridman 1:18:09 Yeah.

Dmitry Korkin 1:18:09 So we need to go back to, to the beginning of our conversation, you know, with the realization that, you know, an average protein is the typical of what, uh, the, the CASP, uh, has been focusing on is, uh, the, you know, the, this competition has been focusing on the single, maybe two domain proteins that are still very compact, and even those ones are extremely challenging to, to solve, right? But now, we talk about, you know, an average protein that has two, three protein domains. If you look at the, um, proteins that, uh, that are in charge of the, you know, uh, of the process, y- you know, with the neural system, right? Uh, perhaps one of the, uh, of the most recently evolved sort of, uh, systems i- i- in, in a, in a organism, right? All of them, well, the majority of them are highly multi-domain proteins. So they are, you know, some of them have five, six, seven, you know, and more domains, right? And, you know, we are very far away from understanding how these proteins are folded.

Lex Fridman 1:19:22 So the complexity of the protein matters here, the complexity, the, the complexity of the protein modules, or the, the protein domains. (sighs) So you're saying, uh, solve, so the definition of solved here is particularly the CASP competition achieving human level, not human level, achieving, uh-

Dmitry Korkin 1:19:42 Experimental level.

Lex Fridman 1:19:43 ... uh, experimental level performance on these particular sets of proteins that have been used in these competitions.

Dmitry Korkin 1:19:50 Well, I mean, you know, I, I, I, I do think that, uh, you know, especially with, with regards to the AlphaFold, you know, it is able to, uh, you know, to solve, you know, at the near experimental level, uh, uh, pretty big majority of the, of the, uh, more compact proteins like, or protein domains, because again, in order to understand how.... the overall protein, uh, uh, you know, mul- multi-domain protein fold, we do need to understand the structure of its individual domains.

Lex Fridman 1:20:28 I mean, unlike if you look at AlphaZero or like even MuZero, if you look at that work, you know, they're... it's nice reinforcement learning, self-playing mechanisms are nice because it's all in simulation, so you can learn from just huge amounts. Like you don't need data. With like the problem with proteins, like the size, uh, I forget how- how much, how many 3D structures have been mapped, but the training data is very small, no matter what. It's like millions, maybe a one or two million, something like that. But some very small number, but like as... it doesn't seem like that's scalable. There has to be... I don't know, it feels like you want to somehow 10X the data or 100X the data somehow.

Dmitry Korkin 1:21:15 Y- yes, but we also can take advantage of, uh, of, uh, homology models, right? So, so the models that are of very good quality because they are essentially, um, obtained based on the evolutionary information.

Lex Fridman 1:21:33 Mm-hmm.

Dmitry Korkin 1:21:33 Right? So, so you can s- there is a potential to enhance this information and, uh, you know, use it again, uh, to, to empower the, the, uh, the training set. Um, and it's... I think I, I am actually very optimistic. I think i- i- it's been one of this, uh, sort of, uh, you know, uh, turning events where you have a system that is, you know, a machine learning system that is truly better than the sort of the more conventional biophysics based methods.

Lex Fridman 1:22:17 That's a huge leap. This is one of those fun questions, but, uh, where would you put it in, in the, uh, ranking of the greatest breakthroughs in artificial intelligence history?

Dmitry Korkin 1:22:30 (laughs)

Lex Fridman 1:22:31 So like, okay, so let's, let's see who's in the running. Maybe you can correct me. So you got like AlphaZero and AlphaGo beating, you know, s- beating the world champion at the game of Go. Thought to be impossible, like 20 years ago, or at least the AI community was highly skeptical. Then you got like also Deep Blue original Kasparov, you have deep learning itself, like the maybe, what would you say, the AlexNet, ImageNet moment, so the first, you know, network ach- achieving human level of performance super, not... that's not true. Achieving like a big leap in performance on the computer vision problem. Uh, there is OpenAI, the whole like GPT-3, that whole space of transformers and language models just achieving this incredible performance, uh, of application of neural networks to language models. Boston Dynamics, pretty cool, like robotics-

Dmitry Korkin 1:23:34 Oh, yeah.

Lex Fridman 1:23:34 ... even though people are like, "There's no AI." Uh, no, no, there's no machine learning currently. But, uh, AI is much bigger than machine learning.

Dmitry Korkin 1:23:44 Yes.

Lex Fridman 1:23:44 So, so that just the engineering aspect, I would say, is one of the greatest accomplishments in engineering side. Engineering, meaning like mechanical engineering of, uh, robotics ever. Then of course, autonomous vehicles, you can argue for Waymo, which is like the Google self-driving car, or you can argue for Tesla, which is like actually being used by hundreds of thousands of people on the road today machine learning system. Um, and, uh, I don't know if you can... What else, what, what else is there? But I think that's it. So, and then AlphaFold many people are saying is up there, potentially number one. Would you put them at number one?

Dmitry Korkin 1:24:24 Well, i- in terms of the impact on, on the science and on the society beyond, it's definitely, you know, to me, would be one of the, you know, uh-

Lex Fridman 1:24:37 Top three. Which you want?

Dmitry Korkin 1:24:39 Maybe. I mean, I'm, I'm-

Lex Fridman 1:24:40 (laughing) Yeah.

Dmitry Korkin 1:24:41 ...pa- pa- probably not the best person to, to, to answer that.

Lex Fridman 1:24:45 (laughing)

Dmitry Korkin 1:24:45 You know, but, you know-

Lex Fridman 1:24:46 Yeah.

Dmitry Korkin 1:24:46 ... I, uh, you know, I, I do have I- I remember my, you know, uh, back in, I think 1997 when Deep Blue beat Kasparov. It was, I mean, it was a shock.

Lex Fridman 1:25:01 Yeah.

Dmitry Korkin 1:25:02 I mean, it was. And I think for the, for the, you know, uh, for the, you know, uh, pretty substantial part of the world that especially people who have some, uh, you know, uh, s- some experience with chess, right? And realizing how incredibly human this game, how, you know, how much of a brain power you need, you know, to, to reach those, you know, those levels of, uh, grandmasters, right, level.

Lex Fridman 1:25:36 And it's probably one of the first time, and how good Kasparov was.

Dmitry Korkin 1:25:39 And, and again, yeah, so K- K- Kasparov's arguably one of the best ever, right?

Lex Fridman 1:25:44 Yeah.

Dmitry Korkin 1:25:45 And you get a machine that beats him, right? So it's, it's in-

Lex Fridman 1:25:48 First time a machine probably beat a human at that scale of a thing, if anything.

Dmitry Korkin 1:25:53 Yes. Yes.

Lex Fridman 1:25:54 Yeah.

Dmitry Korkin 1:25:54 So that was, to me, that was like, you know, one of the groundbreaking events in the history of AI.

Lex Fridman 1:26:00 That's probably number one. As probably like we don't... (laughs) It's, it's hard to remember. It's like Muhammad Ali versus, uh, I don't know, any other, Mike Tyson or something like that. It's like, nah, you got to put Muhammad Ali at number one. Uh, same with, same with Deep Blue even though it's not machine learning based. Uh...

Dmitry Korkin 1:26:19 But still, it, it uses advanced search.

Lex Fridman 1:26:21 Yeah.

Dmitry Korkin 1:26:21 And search is a integral part of AI.

Lex Fridman 1:26:23 Yeah.

Dmitry Korkin 1:26:23 Right? So, so you said, it-

Lex Fridman 1:26:25 It's not... People don't think of it that way now at this moment. In vogue currently, search is not seen as a, as a fundamental aspect of intelligence, but it very well and very likely is.In fact, I mean, that's what neural networks are, is they're just performing search on the space of parameters, and it's all search. (laughs) Uh, all, uh, all of intelligence is some form of search, and you just have to become cleverer and cleverer at that

Lex Fridman 1:26:50 search problem.

Dmitry Korkin 1:26:50 And I also have, uh, another one that you didn't mention that's, that's, that's, uh, one of my favorite ones, is, uh, so y- you've probably heard of this, it's, uh, I think it's called Deep Rembrandt. It's the project where they, they, uh, they trained, uh, I think there was a collaboration between the, uh, sort of the, uh, experts in, in Rembrandt, uh, painting in Netherlands, and a group, an artificial intelligence group, where they train an algorithm to replicate the style-

Lex Fridman 1:27:22 Mm-hmm.

Dmitry Korkin 1:27:22 ... of the Rembrandt, and they actually printed a, a, a, pro- portrait that never existed before-

Lex Fridman 1:27:28 Mm-hmm.

Dmitry Korkin 1:27:28 ... uh, in the style of Rembrandt. They, they, um, I think they printed it on a, on a, sort of, uh, on the canvas that, you know, using pretty much same types of paints and stuff, uh, to me, it was mind-blowing.

Lex Fridman 1:27:44 Yeah.

Dmitry Korkin 1:27:44 It's e-

Lex Fridman 1:27:45 In the space of art, that's interesting. There hasn't been, um, a s- maybe that's, that's it, but I, I think there hasn't been an image net moment yet in the space of art. You haven't been able to achieve superhuman level performance in the space of art, even though there was, you know, s- uh, there was a s- big famous thing where there was pur- a, a piece of art was purchased, I guess, for a lot of money.

Dmitry Korkin 1:28:08 Yes.

Lex Fridman 1:28:09 Yeah.

Dmitry Korkin 1:28:09 Yes.

Lex Fridman 1:28:09 But it's still, you know, people are like, in the space of music at least, um, that's, you know, it's clear that human-created pieces are much more popular. So, there hasn't-

Dmitry Korkin 1:28:22 Yes.

Lex Fridman 1:28:22 ... been a moment where it's like, oh, this is n- we're now... I, I would say in the space of music, what makes a lot of money, we're talking about serious money, it's music and movies or like shows and so on and entertainment. There hasn't been a moment where a- AI created, uh, AI was able to create a piece of music or a piece of, uh, cinema or like Netflix show that is, uh, you know, that's sufficiently popular to make a sh- a ton of money.

Dmitry Korkin 1:28:55 Yeah.

Lex Fridman 1:28:55 And that moment would be very, very powerful 'cause that's like a, that's an AI system being used to make a lot of money and like direct, of course, AI tools, like even Premier, audio editing, a- all the editing, everything I do. To edit this podcast, there's a lot of AI involved. I want, actually there's this, uh, program, I wanna talk to those folks just 'cause I wanna nerd out. It's called Izotope. I don't know if you're familiar with it. They have a bunch of tools of audio processing.

Dmitry Korkin 1:29:20 Mm-hmm.

Lex Fridman 1:29:20 And they have... I think they're Boston-based. Just it's so exciting to be, to, to use it, like, on the audio here, 'cause it's all machine learning. It's not... 'Cause most audi- most audio production stuff is like any kinda processing you do is very basic signal processing, and you're tuning knobs and so on. They have all of that, of course, but they also have all of this machine learning stuff, like where you actually s- give it training data.

Dmitry Korkin 1:29:48 Mm-hmm.

Lex Fridman 1:29:48 You select parts of the audio you train on, you, y- you train on it, and it, it, uh, figures stuff out. It's, it's great. It's able to detect, uh, like the, the ability of it to be able to d- separate voice and music, for example, or voice and anything, is incredible. Like, it, it just, it's clearly exceptionally good at, uh, you know, applying these different neural networks models to, to, to separate the different kinds of signals from the audio. That, that, uh, okay, so that's really exciting. Photoshop, Adobe, people also use it. But to generate a piece of music-

Dmitry Korkin 1:30:27 Yeah.

Lex Fridman 1:30:28 ... that will sell-

Dmitry Korkin 1:30:30 A piece of art.

Lex Fridman 1:30:31 ... millions-

Dmitry Korkin 1:30:31 Yeah.

Lex Fridman 1:30:31 ... a piece of art, yeah.

Dmitry Korkin 1:30:32 Uh, yeah. No, I, I agree, and you know, it's, uh, that's, that's, you know, uh, uh, you know, I, as I mentioned, I offer my, my AI class and, you know, a, an integral part of this is the project, right? So it's, it's my favorite, ultimate favorite part because it, typically we have this, you know, project presentations the last two weeks of the classes right before, you know, the, the Christmas break, and it's, it's sort of, it adds this cool excitement. And every time, I mean, I'm, I'm, I'm amazed, you know, with, with some, uh, some projects that people, uh, you know, come up with. And so, uh, and quite a few of them are actually, you know, they... some- have some link to, uh, to, to arts. I mean, you know, I think last year, uh, we had a group who designed an AI, uh, producing, uh, hokkus, Japanese poems.

Lex Fridman 1:31:28 Oh, wow.

Dmitry Korkin 1:31:29 Uh, so, and some of them-

Lex Fridman 1:31:32 (laughs)

Dmitry Korkin 1:31:32 So, you know, it got trained on the, on the English-based-

Lex Fridman 1:31:35 Are they haikus?

Dmitry Korkin 1:31:36 Haikus, haikus, right, uh... So, um, and, and some of them, you know, they get to present like the, the top selection. They were pretty good.

Lex Fridman 1:31:45 (laughs)

Dmitry Korkin 1:31:45 I mean, you know, I mean, of course, I'm not, I'm not a specialist, but-

Lex Fridman 1:31:48 Yeah.

Dmitry Korkin 1:31:48 ... you, you read them, and you see this rhythm-

Lex Fridman 1:31:50 It seems profound. (laughs)

Dmitry Korkin 1:31:51 Yes, yeah. It, it s- it seems rhythm. So it's, it's kind of cool.

Lex Fridman 1:31:55 (laughs)

Dmitry Korkin 1:31:55 Uh, we, we also had a couple of projects where people tried to, to, uh, teach, uh, AI how to play, like rock music-

Lex Fridman 1:32:03 Mm-hmm.

Dmitry Korkin 1:32:03 ... classical music, uh, I think, and, uh, and, and popular music.

Lex Fridman 1:32:08 Yeah.

Dmitry Korkin 1:32:08 Uh, interestingly enough, uh, you know, classical music was among the most difficult ones.

Lex Fridman 1:32:17 Oh, shit.

Dmitry Korkin 1:32:17 And, and, and you know, of course, if you, if you know, uh, y- uh, you know, if, if you look at the, you know, the, uh, at, at like...... grand masters of music like Bach, right? So there is a lot of, uh, there is a lot of almost math-

Lex Fridman 1:32:34 Yeah. Well, he's very mathematical, right?

Dmitry Korkin 1:32:36 ... yeah, exactly. So- so this is, I would imagine that, yeah, at least some style of this music could be picked up, but then you have this completely different spectrum of- of, you know, classical composers. And so w- you know, it- it- and, you know, i- it's almost like, you know, you don't have to sort of look at the data. You just listen to it and say, "Nah, that's- that's- that's not it. Not yet."

Lex Fridman 1:33:01 That's not it.

Dmitry Korkin 1:33:01 Yeah. Uh-

Lex Fridman 1:33:02 That's- that's how I feel too. There's a, OpenAI has, I think, Open Muse or something like that, the system. It's cool, but it's like, eh, it's not compelling for some- for some reason. It could be a psychological reason too. Maybe we need to have a human being, a tortured soul behind the music. I don't know.

Dmitry Korkin 1:33:20 Yeah. No. That- that... absolutely. I- I completely agree. But yeah, whether or not we'll have, at one day we'll have, you know, a song written by an AI engine to- to be in, like, in top charts-

Lex Fridman 1:33:36 Yeah.

Dmitry Korkin 1:33:36 ... musical charts, I wouldn't be surprised. I wouldn't be surprised.

Lex Fridman 1:33:43 I wonder if we already have one and it just hasn't been announced. (laughs)

Dmitry Korkin 1:33:48 (laughs)

Lex Fridman 1:33:48 We w- we wouldn't know.

Lex Fridman 1:33:50 How hard is the multi-protein folding problem? Is that kind of something you've already mentioned, which is baked into this idea of greater and greater complexity of proteins, like multi-domain proteins? Is that basically become multi-protein, like, complexes?

Dmitry Korkin 1:34:08 So it's- it's... yes. You- you got it right. So- so it's sort of... it has the components of both, of protein folding and protein-protein interactions. Be- uh, because in order for these domains, I mean, s- many of these, uh, proteins actually, they never form a stable structure. Uh, you know, one of my favorite proteins, you