WEBVTT 00:00:03.477 --> 00:00:10.497 This is the Convergent Science Network podcast. Leading researchers in the domain 00:00:10.497 --> 00:00:16.737 of neuroscience, brain theory and technology are interviewed by Paul Verschoor and Tony Prescott. 00:00:21.197 --> 00:00:27.937 This is Paul Verschoor together with Tony Prescott for the Convergent Science 00:00:27.937 --> 00:00:31.917 Network podcast. Here's the 10th edition of the Barcelona Coalition Brain and 00:00:31.917 --> 00:00:35.177 Technology Summer School. And we're here with Ton Koolen. 00:00:36.437 --> 00:00:42.597 And Ton, in your talk this morning, you were dealing with this, 00:00:42.617 --> 00:00:50.537 the analogy or the similarities you might find between neural networks and immune networks. Yeah. 00:00:51.777 --> 00:00:57.217 So how did you come to study or to raise that specific question? 00:00:57.597 --> 00:01:01.257 What kind of mileage do you think we can get in addressing it? 00:01:01.837 --> 00:01:06.257 Yeah, how did I come to it? It was really a case of being asked, 00:01:06.357 --> 00:01:11.877 being lucky to be asked by a group of Italian researchers who had recently developed 00:01:11.877 --> 00:01:13.077 some really nice new models. 00:01:13.757 --> 00:01:18.437 And in the development of these models, they find that at some point, 00:01:18.437 --> 00:01:24.997 solving these models required mathematical technology which had only been in 00:01:24.997 --> 00:01:29.697 the public domain for a couple of years and they were not having the experience 00:01:29.697 --> 00:01:33.757 to do this calculation and I happened to have been working on similar problems. 00:01:35.174 --> 00:01:39.514 And in addition, we knew each other because one of these Italians had done a 00:01:39.514 --> 00:01:43.054 PhD in London. So one thing led to another, and I was asked to join. 00:01:43.854 --> 00:01:47.774 And I really enjoyed that. It was really very interesting. The model was interesting, 00:01:47.914 --> 00:01:55.354 and it allowed me to do research in a field that was really very close mathematically 00:01:55.354 --> 00:01:58.594 to the borderline of what was then the state of the art. 00:01:59.454 --> 00:02:02.074 So this problem had everything you would wish for in a problem. 00:02:02.294 --> 00:02:07.574 It has clear relevance, understanding the immune system has clear medical relevance. 00:02:08.714 --> 00:02:14.734 And it also was very interesting scientifically as someone who likes to do a 00:02:14.734 --> 00:02:17.194 calculation, to be involved in those calculations. 00:02:18.054 --> 00:02:22.074 So what were the specific mathematical techniques that they needed? 00:02:23.134 --> 00:02:27.594 In the analysis of systems with very large numbers of variables, 00:02:27.934 --> 00:02:32.134 the traditional methods that were being used until around 2000. 00:02:33.054 --> 00:02:39.034 Work only when the number of components that each element interacts with is 00:02:39.034 --> 00:02:40.774 either very large or very small. 00:02:41.194 --> 00:02:44.934 But in the intermediate regime where you have components that interact with 00:02:44.934 --> 00:02:49.054 a finite number of other parts, there were no methods available. 00:02:49.274 --> 00:02:52.674 And this is called finite connectivity analysis and there are different mathematical 00:02:52.674 --> 00:02:56.614 medical tricks for dealing with it but they'd only been developed in this domain 00:02:56.614 --> 00:03:04.914 around 2000 by various people including mainly Rémy Monasson in France Giorgio Parisi in Rome. 00:03:06.620 --> 00:03:10.420 And this turned out to be needed to solve these particular immune models. 00:03:10.920 --> 00:03:15.860 And these techniques were already developed looking at specific natural phenomena, 00:03:15.980 --> 00:03:20.780 or these were being developed within theoretical physics or applied mathematics 00:03:20.780 --> 00:03:24.360 just as a technique that people thought was important? 00:03:24.900 --> 00:03:31.100 Well, these methods tend to come from a community of solid-state physicists, 00:03:31.440 --> 00:03:35.300 theoretical physicists, is, but it is specifically the sub-part of that community 00:03:35.300 --> 00:03:37.240 that look at interdisciplinary problems. 00:03:37.480 --> 00:03:41.760 So I think the motivation for looking at finite connectivity problems might 00:03:41.760 --> 00:03:45.880 have even come from optimization problems in computer science that were being 00:03:45.880 --> 00:03:48.520 analyzed by methods from theoretical physics. 00:03:49.440 --> 00:03:54.420 At least the names that come to mind at this moment are typical type of people 00:03:54.420 --> 00:03:56.120 who've been working on optimization problems. 00:03:56.360 --> 00:04:00.420 They convert these optimization problems into theoretical physics problems and 00:04:00.420 --> 00:04:01.160 then use the technology. 00:04:01.960 --> 00:04:06.660 What I find instructive about that is that what you see here is how applied 00:04:06.660 --> 00:04:09.280 problems are driving the basic science. Oh, yes. 00:04:09.520 --> 00:04:12.000 Right? Which is actually… That's very much so. It's counterintuitive. 00:04:12.020 --> 00:04:15.420 Usually, there's this idea like, okay, the basic science happens because we 00:04:15.420 --> 00:04:20.040 have this genius sitting somewhere in a dungeon somewhere, and then it comes 00:04:20.040 --> 00:04:22.700 out and becomes applied. But here, you see it exactly the other way around. 00:04:22.940 --> 00:04:28.880 Yeah. I mean, this has been going on for about 25, 30 years at least in that community. 00:04:28.880 --> 00:04:33.960 Initially, physicists started looking at heterogeneous systems in the context 00:04:33.960 --> 00:04:38.900 of some obscure magnetic materials that later turned out to have no use whatsoever 00:04:38.900 --> 00:04:40.720 anywhere, and were called spin glasses. 00:04:41.760 --> 00:04:46.880 Since then, the main application domains of the mathematical methods that came 00:04:46.880 --> 00:04:52.200 from these obscure materials have been ranging from computer science to economics 00:04:52.200 --> 00:04:54.460 to biology to medicine to statistics. 00:04:56.222 --> 00:05:02.442 And now there is a very fruitful flowing back and forth of ideas and problems 00:05:02.442 --> 00:05:07.582 between theoretical physicists who are only theoretical physicists by virtue 00:05:07.582 --> 00:05:10.542 of the salary they draw from the physics department. 00:05:10.762 --> 00:05:13.602 But they work largely on non-physical problems. 00:05:14.482 --> 00:05:19.122 So this has gone on for quite some time. So one thing I found really remarkable 00:05:19.122 --> 00:05:23.722 in the model you presented from these people in Rome on the immune system in 00:05:23.722 --> 00:05:27.142 this case is that they actually stumbled, if you want, 00:05:27.282 --> 00:05:33.122 into a description of this immune system in terms of an energy landscape, 00:05:33.262 --> 00:05:36.662 if you want, or a dynamical landscape that you could call energy. 00:05:36.662 --> 00:05:43.862 That was identical to what people like John Hopfield have proposed now 30 years 00:05:43.862 --> 00:05:49.442 ago as effective macroscopic descriptions of the spin glass models initially 00:05:49.442 --> 00:05:50.782 and then of neural networks. 00:05:51.782 --> 00:05:57.582 So is this convergence? Was this happenstance, luck, bias because of the techniques 00:05:57.582 --> 00:06:00.342 people had? How did that come together and why is it meaningful? 00:06:01.402 --> 00:06:05.282 Well, I think it's a mixture of many things. First of all, you have to have 00:06:05.282 --> 00:06:11.902 seen certain domains to recognize certain approaches or models or methods for what they are. 00:06:12.442 --> 00:06:18.302 So it demands that you are someone who is willing and able to look over the 00:06:18.302 --> 00:06:19.742 fence away from their own discipline. 00:06:20.042 --> 00:06:24.522 And the people who did this in Rome had been involved in various projects related 00:06:24.522 --> 00:06:27.722 from physical systems to neural networks and so on. 00:06:28.362 --> 00:06:33.522 So while doing the calculations, while building the models, at some point you 00:06:33.522 --> 00:06:38.442 recognize that models are developing into a certain form that resembles something else. 00:06:38.622 --> 00:06:43.342 But you'll only recognize it if you've actually been reading papers or preferably 00:06:43.342 --> 00:06:46.122 been carrying out research in these other disciplines. 00:06:46.542 --> 00:06:51.282 So breadth helps you. And also the other thing you mentioned is mathematical technology. 00:06:51.722 --> 00:06:55.922 Mathematical technology constrains what you will be writing down in the first 00:06:55.922 --> 00:06:56.882 place because Because you know. 00:06:58.257 --> 00:07:03.317 Only certain types of models will be solvable. So you steer away from constructions 00:07:03.317 --> 00:07:05.197 that you know will lead you into the morass anyway. 00:07:05.817 --> 00:07:07.437 So all these things work together. 00:07:09.217 --> 00:07:15.357 From your talk, I understood that both the neural networks and the immune networks 00:07:15.357 --> 00:07:20.937 have this beautiful abstract description in terms of statistical mechanics, 00:07:21.157 --> 00:07:22.437 which you've explored in the talk. 00:07:22.577 --> 00:07:27.297 But I'm interested in, I think you also pointed out in the talk to some other 00:07:27.297 --> 00:07:33.437 sort of direct analogies between immune networks and neural networks that could 00:07:33.437 --> 00:07:37.157 provide further levels of interaction between those two fields. 00:07:37.617 --> 00:07:39.877 Can you say a bit more about what you think those are? 00:07:41.217 --> 00:07:45.197 Yeah, so the similarities are at different levels. 00:07:45.417 --> 00:07:48.597 You have the mathematical level, it was already mentioned just now, 00:07:48.717 --> 00:07:52.277 but also even if you didn't buy these particular models, if you thought, 00:07:52.337 --> 00:07:55.917 well, this is just a choice by a couple of researchers and we could have written 00:07:55.917 --> 00:07:59.957 models in very different ways, although qualitatively perhaps similar. 00:08:00.177 --> 00:08:05.297 But at a higher level you see that you are looking at two systems which share 00:08:05.297 --> 00:08:06.957 various fundamental ingredients. 00:08:07.617 --> 00:08:12.017 A, they have large numbers of variables. B, these variables interact with each 00:08:12.017 --> 00:08:13.337 other in a heterogeneous manner. 00:08:14.077 --> 00:08:17.317 And C, this system is not a static system. 00:08:17.477 --> 00:08:23.357 The forces that all these objects exert onto each other are evolving over time 00:08:23.357 --> 00:08:28.557 on the basis pieces of experience in response to the environment that they work in. 00:08:29.717 --> 00:08:36.197 And these are ingredients that are exactly the same in neural networks and in the immune model. 00:08:36.457 --> 00:08:38.377 There are also differences, that's very clear. 00:08:38.917 --> 00:08:46.277 But the way you can exploit these similarities is first in terms of intuition. 00:08:46.637 --> 00:08:50.217 If you have, as is the case here, one field that has already been studied for 00:08:50.217 --> 00:08:51.517 decades, like neural networks, 00:08:51.737 --> 00:08:57.337 you develop a pretty good intuition for what to expect, pretty good intuition 00:08:57.337 --> 00:09:01.777 for what would happen if you added certain ingredients or took certain ingredients away. 00:09:02.697 --> 00:09:07.177 And that intuition is probably the most powerful thing to carry over to other 00:09:07.177 --> 00:09:09.277 fields, because the details may differ. 00:09:09.517 --> 00:09:14.697 But statistical mechanics teaches us that as soon as you have a very large system, 00:09:14.977 --> 00:09:20.317 microscopic details of how exactly the model is built or what exactly you choose 00:09:20.317 --> 00:09:24.717 for the types of variables or the parameters become less and less important. 00:09:25.217 --> 00:09:30.557 So when at some microscopic level of description what is built into the model 00:09:30.557 --> 00:09:33.317 is similar, then you also expect the behavior to be really similar. 00:09:34.297 --> 00:09:38.897 And that helps you. That helps you in designing models. And there is never a 00:09:38.897 --> 00:09:41.957 perfect model. Every model has a typical shelf life. 00:09:42.937 --> 00:09:48.817 But it helps you in being effective in selecting models, in selecting directions 00:09:48.817 --> 00:09:50.237 in which to to push your models. 00:09:51.956 --> 00:09:57.736 Because at the end of the day, certain models are enormously powerful and helpful, 00:09:57.936 --> 00:10:00.756 and many others will not be so for various reasons. 00:10:01.656 --> 00:10:06.576 And this guides you, this analogy guides you. And I think for some of the neuroscientists 00:10:06.576 --> 00:10:09.456 in the audience, it was also surprising to, 00:10:09.956 --> 00:10:14.696 hear you talking about immune networks and the way that we talk about neural 00:10:14.696 --> 00:10:20.876 networks as structures that have learning and memory, perhaps in quite a different way. 00:10:20.876 --> 00:10:24.236 Way, and in some senses the mechanisms may be similar and other ones might be different. 00:10:24.656 --> 00:10:29.596 But I guess that's one of the paths by which this research in immune networks 00:10:29.596 --> 00:10:31.356 may come back to influence the neuroscience. 00:10:32.076 --> 00:10:36.856 Yes, that is something that is not guaranteed to happen, but it is something 00:10:36.856 --> 00:10:38.916 that I certainly think is worth thinking about. 00:10:40.876 --> 00:10:47.216 I think in science generally it's very rare for one to make an interesting discovery, 00:10:47.216 --> 00:10:48.816 for that discovery not to have 00:10:48.816 --> 00:10:52.196 implications outside of the initial domain where the discovery was made. 00:10:53.156 --> 00:11:00.196 And the main short-term lesson that I think neuroscience can take from this 00:11:00.196 --> 00:11:06.256 immunological way of working is the possibility of having attractor neural networks 00:11:06.256 --> 00:11:11.276 in which one is not recalling one attractor at a time, 00:11:11.336 --> 00:11:13.916 an attractor which contains a very large number 00:11:14.016 --> 00:11:16.716 of bits of information, one image at a time, 00:11:17.176 --> 00:11:22.736 but where you see that there are multiple attractors which can be recalled independently 00:11:22.736 --> 00:11:27.016 at the same time, even though you are looking at a system that is fully interacting, 00:11:27.116 --> 00:11:31.576 not directly, but there are no subsystems that are disconnected from each other. 00:11:32.216 --> 00:11:35.316 But if you then change parameters, all of a sudden you can get a transition 00:11:35.316 --> 00:11:37.856 where all these subsystems are no longer acting independently. 00:11:38.836 --> 00:11:46.756 Now, many of the things that you think about in neuroscience are somewhat reminiscent 00:11:46.756 --> 00:11:53.496 of that, especially in making recognition on the basis of recognizing subparts of images. 00:11:53.836 --> 00:11:59.176 Whether or not you combine them, or whether you want to create bigger objects 00:11:59.176 --> 00:12:03.096 that are composed of things that you've seen before, but never in that combination. 00:12:05.027 --> 00:12:08.067 May be relevant, may not be relevant, but certainly worth exploring. 00:12:09.067 --> 00:12:13.787 These models operate in a slightly different regime in terms of parameters and 00:12:13.787 --> 00:12:16.527 scales and numbers of bits of patterns and things like that. 00:12:17.387 --> 00:12:20.967 One thing I find interesting here is also how you described the development 00:12:20.967 --> 00:12:22.947 of these techniques. We start with spin glasses. 00:12:24.747 --> 00:12:29.507 Those techniques made certain assumptions about symmetry and homogeneity of 00:12:29.507 --> 00:12:34.167 the system. So those techniques would not scale to complex biological phenomena 00:12:34.167 --> 00:12:38.587 like the immune system or nervous systems. For that, you need new techniques. 00:12:39.527 --> 00:12:43.707 Would that actually help us to also distinguish living from non-living systems? 00:12:43.907 --> 00:12:48.287 Because apparently, mathematically, there's a unique set of tools you need to 00:12:48.287 --> 00:12:53.487 describe living systems as opposed to a pure physical system like a spin glass. 00:12:53.807 --> 00:12:58.867 Would you go that far? Yeah, well, one thing I've learned over the years working 00:12:58.867 --> 00:13:03.207 on biological systems is that when it comes to modeling, biology is much harder than physics. 00:13:03.867 --> 00:13:07.187 And there are some really very clear reasons for that. 00:13:07.887 --> 00:13:13.627 One very simple reason is that very often in biology, you look at systems in 00:13:13.627 --> 00:13:20.147 which the number of elements involved is neither really small nor really big, the mesoscopic area. 00:13:20.247 --> 00:13:24.127 And the problem with statistical methods from physics is they only work when 00:13:24.127 --> 00:13:25.027 you have really large numbers. 00:13:25.727 --> 00:13:29.507 In biology, that's not always the case. That's the first obvious thing. 00:13:29.607 --> 00:13:35.707 The second thing is that many physical methods rely on the systems that study going to equilibrium. 00:13:35.987 --> 00:13:40.087 Well, that's never happening in biology, at least not until we are permanently horizontal. 00:13:41.027 --> 00:13:45.247 So that rules out about three quarters of the methodology of statistical physics. 00:13:45.907 --> 00:13:51.647 But I think the main one, and that is also the most interesting one for a modeler 00:13:51.647 --> 00:13:56.767 like like me, is that in biology, we always have heterogeneity. 00:13:56.827 --> 00:14:02.707 We have heterogeneity in interactions between elements, heterogeneity in molecular structures. 00:14:04.649 --> 00:14:08.229 And physics started out, at least statistical mechanics started out, 00:14:08.249 --> 00:14:13.509 looking at systems in which every particle was interchangeable with every other particle. 00:14:14.089 --> 00:14:19.249 Two electrons are exactly the same. The forces between particles would always be the same. 00:14:20.049 --> 00:14:23.669 Physicists tended to look at pure systems, and they saw heterogeneous systems 00:14:23.669 --> 00:14:29.249 as, say, impurities of pure systems, perturbations of pure systems. 00:14:29.449 --> 00:14:35.049 Biology is not like that. Now, in the 70s, there was a whole area opened up 00:14:35.049 --> 00:14:39.869 in statistical mechanics to look at heterogeneous systems. This is where the spin classes came in. 00:14:40.769 --> 00:14:45.809 But one fundamental difference has remained, and that is, in using the physical 00:14:45.809 --> 00:14:52.069 methods, you always have to be able to represent the heterogeneity as randomness. 00:14:52.209 --> 00:14:57.449 You write a probability distribution for the particular parameters or forces 00:14:57.449 --> 00:14:58.349 that are heterogeneous. 00:14:59.289 --> 00:15:04.689 And that's exactly what is different in biology. Biological systems are heterogeneous, 00:15:04.749 --> 00:15:06.849 but the heterogeneity is never random. 00:15:07.249 --> 00:15:11.749 It has always been a consequence, at least partly, of evolution and selection. 00:15:12.649 --> 00:15:17.029 And that rules out immediately all these methods. And the best example to think 00:15:17.029 --> 00:15:19.589 about is probably the modeling of proteins. 00:15:21.115 --> 00:15:26.355 Proteins are these large molecules assembled of little parts that are called amino acids. 00:15:27.335 --> 00:15:32.135 And nature uses proteins as the building blocks of cells, as messengers. 00:15:32.475 --> 00:15:35.235 It's the basic hardware of cellular biology. 00:15:36.235 --> 00:15:42.655 If you look at randomly selected macromolecules of amino acids, 00:15:42.775 --> 00:15:44.775 they behave completely differently from proteins. 00:15:44.775 --> 00:15:50.675 That means that it is essential that, in order to understand how these biological 00:15:50.675 --> 00:15:57.815 systems operate, you must study the evolution that gave rise to these structures. 00:15:58.115 --> 00:16:02.275 But that is a very complicated thing, because it is a two-time-scale process 00:16:02.275 --> 00:16:06.715 in which the processes on the slowest timescale influence the fast ones, 00:16:06.895 --> 00:16:09.055 and the fast ones feed back to evolution. 00:16:10.015 --> 00:16:14.815 You will never be able to write a heterogeneity in terms of a probability distribution. 00:16:15.595 --> 00:16:21.635 That is really interesting because it means that here is a type of problem that 00:16:21.635 --> 00:16:24.975 we cannot handle the way we would handle a heterogeneous physical system. 00:16:25.775 --> 00:16:29.795 There are ways, there are openings, but it is a completely different field. 00:16:30.255 --> 00:16:34.655 You mentioned this in your talk that you were continually discussing 00:16:34.655 --> 00:16:38.275 with the biologists just about which aspects of the system they consider to 00:16:38.275 --> 00:16:44.295 be important and in your models you have a noise term which i guess at the moment 00:16:44.295 --> 00:16:48.335 captures a lot of that detail yeah and there may be aspects of that detail that 00:16:48.335 --> 00:16:51.135 they think has to be pulled out and put into the model, 00:16:52.035 --> 00:16:57.435 is is what you're saying that the mask gets hairier and hairier as you pull 00:16:57.435 --> 00:17:02.055 more of those details out and put into the model or is is there some sort of 00:17:02.055 --> 00:17:04.635 possibility of actually going into 00:17:04.715 --> 00:17:08.495 that space and it's not going to become increasingly difficult. 00:17:08.775 --> 00:17:13.975 There are going to be ways of making models that are more biologically rich, 00:17:14.175 --> 00:17:18.995 sort of things that you can work with. You can still show interesting results. 00:17:19.795 --> 00:17:22.695 I think there's no golden and perfect answer to this. 00:17:25.004 --> 00:17:28.384 It's always the case that in the development of models, you have these three 00:17:28.384 --> 00:17:30.584 distinct stages as I see them. 00:17:30.684 --> 00:17:34.184 So stage one is where you have a problem from the real world and you have really 00:17:34.184 --> 00:17:37.144 no idea how to model it in such a way that you can make progress. 00:17:37.564 --> 00:17:41.424 Now, I think we are beyond that point in immunology now. 00:17:41.624 --> 00:17:46.344 And the second stage is that there are some models that seem to make sense, 00:17:46.424 --> 00:17:51.464 but they are really very simplified, extremely abstract, and one has sacrificed 00:17:51.464 --> 00:17:54.444 a lot of the realism in order to get something that works. 00:17:55.024 --> 00:17:58.204 That's where we are now. And the third stage is, as you say, 00:17:58.244 --> 00:18:00.404 where you make these models more realistic. 00:18:01.204 --> 00:18:07.724 And that means having a good intuition for knowing in which order to bring these ingredients in. 00:18:07.824 --> 00:18:12.464 And there will be some ingredients that have a serious impact on the methodology 00:18:12.464 --> 00:18:14.524 that you will need to still solve the models. 00:18:15.404 --> 00:18:19.064 And there are some ingredients that are relatively easily incorporated. 00:18:19.084 --> 00:18:24.284 To give an example in these immune models, the initial models of Adriano and Elena, 00:18:25.104 --> 00:18:31.444 had the unwelcome element that a T-cell, a single T-clone, could be simultaneously 00:18:31.444 --> 00:18:35.244 excitatory to some B-cell and inhibitory to another. 00:18:35.464 --> 00:18:39.444 It's just like, and it's a nice analogy, the same thing was assumed in neural 00:18:39.444 --> 00:18:41.444 network theory in the very beginning by Hopfield. 00:18:43.116 --> 00:18:47.056 We know that the hardware is different. The hardware doesn't allow for that. 00:18:47.196 --> 00:18:52.296 And in neural networks, we could remove this and we could still solve models. The same happened here. 00:18:52.436 --> 00:18:56.556 So that happened to be a relatively easy ingredient to bring in. 00:18:57.356 --> 00:19:00.116 There are other things that are going to be much harder. We can see that coming. 00:19:00.216 --> 00:19:03.276 And that is when you include the evolution of antigen. 00:19:03.416 --> 00:19:09.116 Because then you have to study in great detail the dynamical processes of hypermutation. 00:19:09.116 --> 00:19:16.296 Mutation, because it is a rat race between the virus or the bacterium that has 00:19:16.296 --> 00:19:20.796 come in and that will multiply, and the immune system trying to catch up. 00:19:21.356 --> 00:19:24.796 So when you start working on those models, you can forget about equilibrium 00:19:24.796 --> 00:19:28.296 studies, you can forget about energy functions, because these systems are no 00:19:28.296 --> 00:19:29.216 longer going to equilibrium. 00:19:29.636 --> 00:19:33.656 So you're going to move away from statistical mechanics to another modelling approach? 00:19:33.876 --> 00:19:36.736 Well, one would move away from equilibrium statistical mechanics, which 00:19:36.736 --> 00:19:39.796 is the part that has been used so far on these particular models 00:19:39.796 --> 00:19:42.716 then you would use non-equilibrium statistical mechanics and 00:19:42.716 --> 00:19:46.776 that is a slightly more specialist area 00:19:46.776 --> 00:19:54.156 of statistical mechanics it is a small subset of the broader area most of statistical 00:19:54.156 --> 00:19:59.796 mechanics but this non-equilibrium system mechanics has been developing by looking 00:19:59.796 --> 00:20:03.996 at biology or it came from other domains of physics oh it came also from from 00:20:03.996 --> 00:20:05.256 standard domains of physics, 00:20:05.376 --> 00:20:08.776 because even if you know what the equilibrium behavior is, you might still be 00:20:08.776 --> 00:20:11.336 interested in studying how the system would get there. 00:20:11.676 --> 00:20:15.876 So they were developed already within the physics community, 00:20:16.136 --> 00:20:20.556 but you just find that in biology, that will turn out to be your only toolbox, 00:20:20.876 --> 00:20:23.416 because most biological systems don't go to equilibrium. 00:20:24.256 --> 00:20:30.436 Okay, the analogy between neural systems, biological neural systems, 00:20:30.496 --> 00:20:34.236 and immune immune systems is that they can learn and have memory, right? 00:20:37.211 --> 00:20:40.811 So, in the nervous system, and there's also many talks on this, 00:20:40.971 --> 00:20:45.991 the dominant view on that is that this memory has something to do with the connectivity 00:20:45.991 --> 00:20:47.391 in the network and how it's regulated. 00:20:48.011 --> 00:20:49.811 It might not be literally the 00:20:49.811 --> 00:20:52.711 synapse, but it is the regulation of the synapse that's key for memory. 00:20:53.431 --> 00:20:57.391 How should we think about memory now in this immune system? What's the substrate 00:20:57.391 --> 00:20:58.871 of memory in the immune system? 00:20:59.871 --> 00:21:04.091 Yeah, that has been a very tricky discussion in the biology community in the 00:21:04.091 --> 00:21:08.111 sense that there are various competing views. 00:21:08.451 --> 00:21:13.091 Some are fashionable in certain eras and then get ditched a couple of years 00:21:13.091 --> 00:21:15.051 later, and that has happened in this community. 00:21:15.151 --> 00:21:22.111 For instance, Jern's model of idiotic networks envisaged a mechanism of memory 00:21:22.111 --> 00:21:26.591 that was very different from what most biologists now assume is the case. 00:21:26.591 --> 00:21:30.811 Most biologists now assume that bee clones, after they've been activated, 00:21:30.991 --> 00:21:36.051 after they've been producing antibodies, then go into a state where they become 00:21:36.051 --> 00:21:37.091 so-called memory cells. 00:21:37.991 --> 00:21:44.431 So somehow you want to retain and memorize the particular shapes of antibodies 00:21:44.431 --> 00:21:47.851 that were very successful in attacking the antigen that came in. 00:21:48.031 --> 00:21:51.311 So the thinking is that those B cells that produce those antibodies, 00:21:51.691 --> 00:21:59.011 they convert into a state where they can live for 25 years. I find that intuitively slightly. 00:22:01.319 --> 00:22:07.959 Not maybe unrealistic, but slightly far-fetched. Is there any empirical support for that idea? 00:22:08.579 --> 00:22:13.699 Well, I'm not an immune biologist, and I'm sure there is some support. 00:22:13.919 --> 00:22:22.179 But I would be surprised if that was a clear-cut and completely closed case. 00:22:22.419 --> 00:22:26.359 I think there is still room for alternative. But that would then predict that 00:22:26.359 --> 00:22:30.339 there is some reservoir of these memory B-cells that reside somewhere in the 00:22:30.339 --> 00:22:31.699 body. Yes, that's the problem. 00:22:31.879 --> 00:22:33.939 That's expanding, right? That's called ditty expanding. Well, 00:22:33.939 --> 00:22:35.179 the thinking is they don't expand. 00:22:35.319 --> 00:22:38.699 The thinking is that you have this group of B cells that have been activated, 00:22:38.799 --> 00:22:44.079 and they go and sit somewhere, and they live 30 years, just in case at some 00:22:44.079 --> 00:22:45.879 point that particular virus reappears. 00:22:47.339 --> 00:22:52.259 I don't buy that. But I cannot argue that there is evidence to reject this. 00:22:52.379 --> 00:22:54.879 It just seems a bit far-fetched to me. Okay, but it means for you, 00:22:54.959 --> 00:22:59.279 this means the memory problem for immune systems is still not resolved. 00:22:59.279 --> 00:23:00.579 If we don't actually really know? 00:23:01.199 --> 00:23:07.459 I think there will be an explanation which is a bit more natural and elegant than the memory cell. 00:23:07.659 --> 00:23:14.059 So your modeling approach, would that lend itself to building models that contest 00:23:14.059 --> 00:23:15.039 the alternative hypothesis? 00:23:16.159 --> 00:23:19.999 I think so. Yes, I think so. I think so. But at this moment, 00:23:20.059 --> 00:23:23.319 we're not there yet because the models are relatively primitive. 00:23:23.959 --> 00:23:26.919 I mean, compared to, if you look at, that's why I find the history of neural 00:23:26.919 --> 00:23:30.239 networks here is so interesting because we've gone through many of those things. 00:23:30.339 --> 00:23:33.379 And you can look back and see, yeah, it was very similar then. 00:23:33.799 --> 00:23:37.379 Hopfield models were primitive, really very primitive. Lutz was wrong with them. 00:23:39.561 --> 00:23:43.061 And the interesting thing was, at that time even, Hopfield models were still 00:23:43.061 --> 00:23:46.881 solved using equilibrium statistical mechanics, and in these models, 00:23:46.981 --> 00:23:50.261 one sacrificed biological realism in order to be able to do that. 00:23:51.121 --> 00:23:55.401 And when you talked to people in those days, it was very often suggested, 00:23:55.641 --> 00:23:59.341 well, we will never be able to do the dynamics, so this is why we do it that way. 00:23:59.701 --> 00:24:03.601 Turns out to be not true. People are sometimes too pessimistic about what is 00:24:03.601 --> 00:24:08.061 possible, and that is very dangerous, because you miss opportunity. 00:24:08.061 --> 00:24:09.561 Turned out to be nonsense. 00:24:09.941 --> 00:24:14.641 Doing the dynamics was, in retrospect, relatively easy, which was very good 00:24:14.641 --> 00:24:18.161 because then we could remove all these biologically unrealistic model ingredients. 00:24:18.641 --> 00:24:20.321 I expect the same to happen here. 00:24:21.121 --> 00:24:25.281 But the other thing is that for memory, it's mysterious. 00:24:26.861 --> 00:24:30.901 In some sense, we could argue memory is equally mysterious for the brain, 00:24:31.741 --> 00:24:36.161 but there is consensus on what that substrate would be, more or less. 00:24:36.161 --> 00:24:38.841 But now let's look at the learning component, right? 00:24:38.901 --> 00:24:44.161 So for the immune system, you would have hypermutation and then selection upon 00:24:44.161 --> 00:24:50.061 this set of hypermutated B cells as your core learning mechanism, right? 00:24:51.441 --> 00:24:57.441 And now in the brain, yes, people like Jerry Edelman have proposed that similar 00:24:57.441 --> 00:25:02.961 mechanisms would work in the brain to select synapses and in that way have specificity 00:25:02.961 --> 00:25:04.281 of connectivity and learning. 00:25:05.081 --> 00:25:09.501 But it's not the only way in which we can think about learning in the nervous system. 00:25:09.541 --> 00:25:15.881 It might also just be the differential strengthening of synapses that allow 00:25:15.881 --> 00:25:20.101 you, as in a Hopfield network, right, to sort of bias the network towards certain 00:25:20.101 --> 00:25:24.221 responses away from other responses without necessarily selecting across a repertoire. 00:25:24.241 --> 00:25:27.001 You remove certain possibilities completely. Completely. 00:25:28.401 --> 00:25:36.041 So do you believe that selection is then the only mode of learning in the immune network? 00:25:36.261 --> 00:25:38.801 Or is there actually more subtle than that? 00:25:40.001 --> 00:25:45.421 It's an interesting question. I think that over the years we find that a lot 00:25:45.421 --> 00:25:48.001 more is going on than we presently think. 00:25:48.121 --> 00:25:52.161 I would also, for instance, expect that at some point there is some interface 00:25:52.161 --> 00:25:57.461 between the nervous system and the immune system. And even today I had some 00:25:57.461 --> 00:26:01.901 very nice discussions with some of the people here where these ideas came up too. 00:26:02.761 --> 00:26:08.321 If the nervous system, for instance, would be able to predict or anticipate 00:26:08.321 --> 00:26:12.601 certain invasions of antigen, then you can prepare the immune system. 00:26:15.723 --> 00:26:19.403 I think there's a lot going to be found that we don't know present. 00:26:20.663 --> 00:26:25.223 Well, in that sense, for instance, if you look at sepsis as an inflammatory 00:26:25.223 --> 00:26:28.603 problem, it's extremely interesting to look at. 00:26:28.623 --> 00:26:32.883 If you look at these broad organ systems that are affected in sepsis by inflammation, 00:26:33.683 --> 00:26:39.263 it's a rather complex dynamics in which, let's say, the inflammatory load to 00:26:39.263 --> 00:26:46.683 organs can be very specifically changing or transiently changing during such a disease process. 00:26:47.563 --> 00:26:52.083 And actually, also many patients who suffer from sepsis, some of them might 00:26:52.083 --> 00:26:56.223 appear in all standard biomarkers healthy. 00:26:56.383 --> 00:26:59.703 They're sent home, and a few months later, they still die, right? 00:26:59.843 --> 00:27:05.023 So that would also suggest there's much more active regulation going on of the global response, 00:27:05.023 --> 00:27:08.143 response yeah with the possible anticipatory component to 00:27:08.143 --> 00:27:11.823 the immune response that might or 00:27:11.823 --> 00:27:15.123 might not reside in the immune system itself so 00:27:15.123 --> 00:27:18.043 how do you see that coupling exactly between the brain and 00:27:18.043 --> 00:27:21.583 the immune system do you expect the brain to really model let's say the health 00:27:21.583 --> 00:27:28.963 status of organ systems and then bias in some way the immune response and if 00:27:28.963 --> 00:27:32.863 so what's the substrate of that you have to think really about these things 00:27:32.863 --> 00:27:35.583 in in terms of the coding that the different system use. 00:27:35.743 --> 00:27:40.903 Brains think in terms of electric potentials and the immune system thinks in terms of chemistry. 00:27:41.423 --> 00:27:46.843 It communicates by docking and sending cytokines and things like that. 00:27:47.703 --> 00:27:53.823 And there are certain constraints if you think about possible mechanisms and possible interfaces. 00:27:53.923 --> 00:27:57.623 For instance, immune cells cannot pass the blood-brain barrier, 00:27:57.763 --> 00:27:59.923 so they cannot directly enter the brain. 00:28:00.723 --> 00:28:03.723 However, various molecules that they produce could. 00:28:04.643 --> 00:28:10.363 We also know that in the brain, as I understand it, there is an interface between 00:28:10.363 --> 00:28:14.143 communication by spikes and chemistry. 00:28:14.283 --> 00:28:18.723 The release of all kinds of chemicals happens also in the brain. 00:28:19.583 --> 00:28:22.603 So I don't think it's far-fetched. I don't think it's far-fetched. 00:28:22.703 --> 00:28:25.983 There are all kinds of opportunities, but we will never be able to find out 00:28:25.983 --> 00:28:32.003 until we have at least a satisfactory description of the immune system itself. 00:28:32.763 --> 00:28:34.923 There's so much we still don't know about the immune system. 00:28:35.283 --> 00:28:38.383 Going to an interface now is premature. 00:28:39.603 --> 00:28:43.143 But if I were a betting person, I would put my money on there being one. 00:28:43.503 --> 00:28:47.643 It just seems to be giving you an enormous evolutionary advantage. 00:28:49.043 --> 00:28:51.103 And therefore I think it will be there. 00:28:53.020 --> 00:28:56.560 So the algorithms that are involved in learning, presumably, 00:28:56.900 --> 00:28:59.300 I mean, you didn't talk about them in any detail in the talk, 00:28:59.400 --> 00:29:04.140 but there's something analogous happening there to natural selection, 00:29:04.340 --> 00:29:10.000 and is it involving both the B-cell and the T-cell populations in some interesting 00:29:10.000 --> 00:29:15.380 ways that alters the population dynamics so as to give you some form of optimal search? 00:29:15.660 --> 00:29:21.880 Yeah, so it's interesting if you look at the literature on immunology that most 00:29:21.880 --> 00:29:26.120 studies that have been carried out tend to have been carried out on B-cells 00:29:26.120 --> 00:29:29.420 and on evolution of receptors of B-cells. 00:29:29.420 --> 00:29:32.220 A lot is known about them, much less is known about T-cells. 00:29:32.220 --> 00:29:37.040 And partially this is due to the fact that T-cells were only discovered later. 00:29:37.700 --> 00:29:41.380 The regulatory T-cells were only discovered in the 1990s. 00:29:41.920 --> 00:29:46.120 So there is a good reason why we know much more about B-cells than we know about T-cells. 00:29:46.620 --> 00:29:50.940 When you think about learning, and this is This is again where I find these 00:29:50.940 --> 00:29:53.420 models very helpful, the models 00:29:53.420 --> 00:29:56.340 that create this mapping between neural network and the immune system. 00:29:56.860 --> 00:30:01.160 If you look at those models, you see where is the learning residing, 00:30:01.300 --> 00:30:05.460 what is the substrate for the learning? It is in the cytokine variable. 00:30:06.160 --> 00:30:11.400 So if you think about what they do, the cytokine variables simply tell you exactly 00:30:11.400 --> 00:30:15.420 which B clone talks to which T clone, and within these models, 00:30:15.540 --> 00:30:16.720 it is also in the other direction. 00:30:18.560 --> 00:30:23.320 And that is exactly what is being changed as a result of hypermutation and selection, 00:30:23.980 --> 00:30:25.780 when we go through that process. 00:30:26.080 --> 00:30:29.900 Because if we change the receptor properties of the Bs and the Ts, 00:30:30.140 --> 00:30:32.240 then we also change who talks to whom. 00:30:32.680 --> 00:30:37.880 So we're effectively changing the patterns that, in the language of neural networks, 00:30:38.100 --> 00:30:41.620 would be the patterns stored in a Hopfian model. So we're rewiring the network? 00:30:41.980 --> 00:30:47.240 We are rewiring the network. At the same time as we're changing the specificity of the cells? else. 00:30:47.300 --> 00:30:51.680 So in a way, the network is a consequence of the information being stored. 00:30:51.880 --> 00:30:55.760 And since we changed the information being stored, we're effectively rewiring the network. 00:30:56.120 --> 00:31:01.420 So we could see the cytokine as a virtual synapse now that connects certain... No, not the synapse. 00:31:01.560 --> 00:31:04.400 The synapse is expressed in terms of the cytokines. 00:31:04.580 --> 00:31:07.200 So the cytokines would be... They're the connection. 00:31:07.420 --> 00:31:10.240 They form the connection between the B's and the T's. They are the equivalent 00:31:10.240 --> 00:31:12.360 thing of the bits in the patterns of the Hopfield model. 00:31:12.620 --> 00:31:14.920 Exactly. And they build the synapse. 00:31:15.840 --> 00:31:20.780 So what So what's now the capacity of then such a memory? 00:31:21.000 --> 00:31:24.260 If the cytokines would then, in this, the analogy we're now pursuing, 00:31:24.580 --> 00:31:27.720 the cytokines are in that your connectivity matrix, right? 00:31:28.800 --> 00:31:32.320 So given the scale of the immune system and the number of elements, 00:31:32.440 --> 00:31:35.580 what kind of capacity would we then look at in these cytokines? 00:31:35.620 --> 00:31:39.980 And what kind of learning rates would you expect or the rate of change that you can support? 00:31:40.280 --> 00:31:45.280 Yeah, this we can now answer as a result of the models. And the answer is, in fact, very trivial. 00:31:45.280 --> 00:31:48.200 This is always when you have understood something you 00:31:48.200 --> 00:31:52.300 know that you could have thought about it before but in standard 00:31:52.300 --> 00:31:55.260 hopfield models what you know is that you can store if you 00:31:55.260 --> 00:31:59.140 want to have perfect recall only relatively small number of patterns but each 00:31:59.140 --> 00:32:02.740 of these patterns has an extensive number of bits here is the other way around 00:32:02.740 --> 00:32:07.560 you you can store and recall an extensive number of patterns proportions to 00:32:07.560 --> 00:32:11.720 a number of b clones but each of them will have only a finite that number of 00:32:11.720 --> 00:32:12.920 bits of information in that. 00:32:14.020 --> 00:32:20.260 So the capacity is in the order of n, and precise numbers you find by calculating 00:32:20.260 --> 00:32:22.800 the phase diagrams, which has been possible now. 00:32:23.300 --> 00:32:29.940 You can compute these things. Right. But then another kind of connectivity matrix 00:32:29.940 --> 00:32:33.820 might be defined by the T helper cells, for instance, that would modulate the 00:32:33.820 --> 00:32:35.200 interaction between the Ts and the Bs. 00:32:36.940 --> 00:32:39.280 For the nervous. 00:32:41.336 --> 00:32:46.576 Just focus the whole learning problem towards the synapses keeps things relatively 00:32:46.576 --> 00:32:47.676 controllable, understandable. 00:32:48.276 --> 00:32:52.056 But maybe in the immune system, there might be multiple levels at which this 00:32:52.056 --> 00:32:55.636 plasticity now gets expressed or multiple substrates or not. 00:32:55.716 --> 00:33:01.916 Are you happy with just focusing it fully for now at this cytokine modulation? 00:33:02.916 --> 00:33:07.696 I think for now, this is what is being done simply because we are already very 00:33:07.696 --> 00:33:09.776 happy with what we can achieve with this model. 00:33:09.776 --> 00:33:13.016 But in the future we 00:33:13.016 --> 00:33:16.196 will want to study in more detail the actual process 00:33:16.196 --> 00:33:18.916 of what you would say learning in neural networks what we 00:33:18.916 --> 00:33:21.916 would call hypermutation and thereby the changing 00:33:21.916 --> 00:33:25.596 of all these cytokine variables in the immune system and 00:33:25.596 --> 00:33:28.296 then we will undoubtedly find that things get more 00:33:28.296 --> 00:33:31.176 complicated we will have to look at different types 00:33:31.176 --> 00:33:33.856 of p cells different types of t cells and all those 00:33:33.856 --> 00:33:36.756 things so you will inevitably get a 00:33:36.756 --> 00:33:39.656 next layer of complexity simply because you move away 00:33:39.656 --> 00:33:42.796 from what is currently one of the most primitive starting 00:33:42.796 --> 00:33:45.696 points imaginable so that 00:33:45.696 --> 00:33:49.596 will happen that will happen but um i 00:33:49.596 --> 00:33:55.776 think on top of that we will find and that is going to be done by experimentalists 00:33:55.776 --> 00:33:59.536 we will probably find new players that we haven't even been thinking of not 00:33:59.536 --> 00:34:04.736 just b cells or t cells but other cells that do things we might find interactions 00:34:04.736 --> 00:34:06.656 mechanisms Mechanisms that we don't know now. 00:34:08.196 --> 00:34:11.196 But now think about the signaling. So in the nervous system, 00:34:11.276 --> 00:34:13.836 in the end, it's electrical, as we discussed. 00:34:14.656 --> 00:34:19.276 In the immune system, it could be through the metabolites, molecules that are 00:34:19.276 --> 00:34:21.236 going through the circulatory systems. 00:34:23.676 --> 00:34:28.156 What are the forms of signaling that now the immune system has available that 00:34:28.156 --> 00:34:31.936 would make it different from the kind of signaling you would see in a nervous system? 00:34:34.176 --> 00:34:41.376 Um let me think about that one so in the nervous system i think there's one, 00:34:42.476 --> 00:34:46.816 complication that maybe is not so much a complication in the immune system and 00:34:46.816 --> 00:34:52.916 that is in the nervous system we also have topology we have wiring and at least 00:34:52.916 --> 00:34:55.696 in the immune system we don't have that because these cells just move around 00:34:55.696 --> 00:34:57.376 in the blood the assumption is that 00:34:57.436 --> 00:35:02.996 there are no barriers that single out certain B-cells as being having given 00:35:02.996 --> 00:35:05.416 access to certain areas as opposed to others. 00:35:05.536 --> 00:35:11.316 There are some aspects only which are that the hyposomatic mutation takes place 00:35:11.316 --> 00:35:13.196 in certain parts of the body. 00:35:13.276 --> 00:35:16.916 But apart from that, we have no topology and that makes your life a lot easier. 00:35:17.236 --> 00:35:22.196 So in terms of the signaling, there is the extra constraint in neural network 00:35:22.196 --> 00:35:25.716 systems that you have to look carefully at the wiring. 00:35:26.376 --> 00:35:30.016 These things cannot float around, neurons cannot float around and establish 00:35:30.016 --> 00:35:31.796 and interact with whoever they want. 00:35:33.072 --> 00:35:39.772 And space is a tricky thing, and especially if you think about mathematical methods from physics. 00:35:40.692 --> 00:35:45.052 In one area, and that is space, they are remarkably unsuccessful. 00:35:45.992 --> 00:35:50.912 And that is when you look, for instance, at a very simple cubic lattice of molecules 00:35:50.912 --> 00:35:54.272 that only talk to their neighbors, that is still an unsolved problem. 00:35:55.152 --> 00:35:58.692 Even if these variables are just binary, they can be on or off, 00:35:58.732 --> 00:36:00.892 and they talk to their neighbors in a three-dimensional arrangement, 00:36:01.192 --> 00:36:04.432 it's an unsolved problem so space brings an 00:36:04.432 --> 00:36:07.592 extra dimension that makes problems really 00:36:07.592 --> 00:36:10.592 significantly harder and that is 00:36:10.592 --> 00:36:13.592 absent in the immune system so that is a difference which actually works 00:36:13.592 --> 00:36:17.512 to the advantage of modeling in the immune system and 00:36:17.512 --> 00:36:20.172 in brains we can get away with it we have been able to 00:36:20.172 --> 00:36:23.092 get away with it for a long time because at least neurons 00:36:23.092 --> 00:36:26.592 talk to a great number of other neurons if neurons 00:36:26.592 --> 00:36:29.852 had only been talking to neighboring neurons we would 00:36:29.852 --> 00:36:32.552 be in real trouble and thank god we're 00:36:32.552 --> 00:36:35.332 not right so then 00:36:35.332 --> 00:36:41.812 but how can we now test these models because um also look at the case of neuroscience 00:36:41.812 --> 00:36:46.332 right there was a lot of enthusiasm for the hopfield kind of networks because 00:36:46.332 --> 00:36:50.472 people had this this hope like okay now it becomes mathematically tractable 00:36:50.472 --> 00:36:54.772 neuroscience can become like physics we can have analytic solutions god knows what, 00:36:56.332 --> 00:37:00.152 but that was 85 more or less, right? So we're talking about 30 years. 00:37:01.812 --> 00:37:05.732 And has that impact been so overwhelming? Are we really now radically, 00:37:05.932 --> 00:37:08.372 have we radically changed the way which we think about the brain? 00:37:08.932 --> 00:37:13.692 Not so sure about that, right? So on the other hand, it has been very, 00:37:13.752 --> 00:37:16.452 as a heuristic, it has helped people to pose certain questions. 00:37:17.292 --> 00:37:20.372 So now in something you can argue for the immune system, you stand now at the 00:37:20.372 --> 00:37:23.372 beginning of an equivalent process, right? 00:37:23.372 --> 00:37:28.492 Where it is this fundamental insight using these techniques from theoretical 00:37:28.492 --> 00:37:32.732 physics has helped you to get the handle on a system that you don't understand. 00:37:33.632 --> 00:37:37.592 But what's the trajectory that you now envision? What are the next steps to 00:37:37.592 --> 00:37:39.552 also make it empirically testable? 00:37:39.952 --> 00:37:46.752 Yeah, so we are very aware of the danger of the same thing happening in immune 00:37:46.752 --> 00:37:48.232 modeling and what happened in neural networks. 00:37:48.332 --> 00:37:52.192 And we can come back to what I think are the reasons for that because some of 00:37:52.192 --> 00:37:53.712 them are not scientific but psychological. 00:37:54.632 --> 00:37:57.912 Now at the moment what we're trying to do is we're trying to, 00:37:58.012 --> 00:38:03.752 in a particular multidisciplinary project that we've started in the UK involving 00:38:03.752 --> 00:38:09.552 not just people like me who do models but also bioinformaticians and experimental immunologists, 00:38:09.852 --> 00:38:16.512 we're trying to find indeed models in which we can test the predictions of the 00:38:16.512 --> 00:38:17.652 model using experiments. 00:38:18.072 --> 00:38:24.112 So we would do this in the form of having a model in which an antigen comes 00:38:24.112 --> 00:38:31.312 in that has never been seen before, and then predicting the dynamical response, first and secondary. 00:38:33.432 --> 00:38:38.872 And see to what extent this is similar to what we get in what is called unchallenged 00:38:38.872 --> 00:38:42.032 individuals in an immunological response. 00:38:43.252 --> 00:38:48.672 And then try to build the models from there, test every time that we can make a prediction. 00:38:49.132 --> 00:38:53.692 At the end of the day, the only real test of whether you have extracted knowledge 00:38:53.692 --> 00:38:58.372 in any area of modeling is whether you can predict unseen situations with the 00:38:58.372 --> 00:39:01.572 model and and then test them. If you can't, you haven't learned anything. 00:39:02.072 --> 00:39:07.092 If you can, you have. And what's the status of the measuring techniques that 00:39:07.092 --> 00:39:11.752 the immunologists have to detect this sort of... 00:39:11.752 --> 00:39:15.852 I mean, for your models, you're looking for quite a lot of detail in terms of 00:39:15.852 --> 00:39:20.092 the temporal and spatial dynamics, or certainly temporal dynamics of what's unfolding. 00:39:20.232 --> 00:39:23.952 So this is really difficult, in the sense that it is very hard, 00:39:24.092 --> 00:39:28.512 and we do struggle with that, to find people with 00:39:28.512 --> 00:39:31.812 the right type of data because you're looking for individuals 00:39:31.812 --> 00:39:35.032 who are who've never 00:39:35.032 --> 00:39:38.472 experienced challenged by a particular antigen before so 00:39:38.472 --> 00:39:41.172 you have to go for really rare diseases you cannot go for 00:39:41.172 --> 00:39:45.332 flus or things like that because there is almost no one who will have a primary 00:39:45.332 --> 00:39:50.352 response to that so you have to go for very funny things like yellow fever and 00:39:50.352 --> 00:39:56.872 And our collaborators are trying to get these people from populations of individuals 00:39:56.872 --> 00:40:00.172 who are about to make a trip abroad to an area where they've never been, 00:40:00.312 --> 00:40:02.732 and who are then being inoculated beforehand. 00:40:03.212 --> 00:40:06.292 And these are the ones that we try to get the data from, because then they get 00:40:06.292 --> 00:40:12.972 a response to this inoculation that should be very similar to primary response. 00:40:12.972 --> 00:40:18.372 Response, because then at least we can do tests before having modeled the secondary 00:40:18.372 --> 00:40:21.612 response, before having modeled the consequences of the hypermutation. 00:40:22.272 --> 00:40:24.012 But why no animal models? 00:40:25.452 --> 00:40:30.632 Well, why no animal models? It just happens to be that the collaborators we 00:40:30.632 --> 00:40:31.952 have don't work with animals. 00:40:32.172 --> 00:40:36.272 They work with people, because they come from a clinical environment where it's 00:40:36.272 --> 00:40:42.432 human immunology they work on. And can you do some of this in a sort of petri 00:40:42.432 --> 00:40:47.012 dish where you just put some antigen into a sample of blood? 00:40:47.332 --> 00:40:49.952 I've been told that this is really hard. Right, okay. 00:40:50.712 --> 00:40:55.752 That it is one of the tricky aspects of experimental immunology. 00:40:56.892 --> 00:41:01.972 Look, here we have all these Brits, they're going on holidays to faraway exotic 00:41:01.972 --> 00:41:07.932 places where they're going to run the risk to catch all sorts of strange diseases 00:41:07.932 --> 00:41:11.012 and and they get inoculated, and imagine this whole thing now works, 00:41:11.752 --> 00:41:14.072 what's the specific prediction you would like to see tested first? 00:41:15.563 --> 00:41:21.523 We would like to see the evolution in time of the distribution of clone sizes. 00:41:21.763 --> 00:41:26.283 So if immune system is not challenged, so let's say before these people get 00:41:26.283 --> 00:41:29.983 the inoculation, the theory can do a prediction already on the distribution 00:41:29.983 --> 00:41:31.623 of clone sizes in normal state. 00:41:32.423 --> 00:41:37.903 Then comes a challenge. This challenge has a consequence, and you see this in 00:41:37.903 --> 00:41:40.283 the distribution of clone sizes. 00:41:40.383 --> 00:41:44.203 And the distribution of clone sizes is a typical thing that can easily be measured. 00:41:44.203 --> 00:41:47.923 It but this is the first test it doesn't mean that this is the perfect test 00:41:47.923 --> 00:41:53.183 just means this is a very clear transparent test and if it doesn't even pass 00:41:53.183 --> 00:41:56.383 that test then you know you have missed something important in your model you 00:41:56.383 --> 00:41:58.823 go back to the drawing board and you think what that could be, 00:41:59.603 --> 00:42:04.763 but this is relatively accessible data and this is what we're focusing on so 00:42:04.763 --> 00:42:08.343 it's a distribution of the sizes of the clones but it's like fine-tuning the 00:42:08.343 --> 00:42:10.003 model is the model falsifiable 00:42:10.223 --> 00:42:13.003 so what would be the observation in this experiment that 00:42:13.003 --> 00:42:15.643 would completely kill the model that you know 00:42:15.643 --> 00:42:18.483 proposed with your colleagues well i don't 00:42:18.483 --> 00:42:21.243 know what the definition of completely kill is because you can always go back 00:42:21.243 --> 00:42:25.043 to the drawing board well that's the question right yeah exactly so but it this 00:42:25.043 --> 00:42:29.723 is a falsifiable model in the sense that you could find a clone size distribution 00:42:29.723 --> 00:42:34.403 which is qualitatively very different from what the model predicts and it's 00:42:34.403 --> 00:42:38.023 not something that you could fix with matching parameters and things like that because. 00:42:39.103 --> 00:42:44.523 So it could work it could not work it is a falsifiable thing but but. 00:42:45.919 --> 00:42:49.719 The tricky area is the area in the middle, where it works to some extent. 00:42:50.519 --> 00:42:53.739 It is not complete nonsense that comes out. It is also not perfect. 00:42:53.859 --> 00:42:57.659 This is the general situation we find ourselves in, in modeling biological systems. 00:42:58.419 --> 00:43:05.879 So then you try to systematically think about the ingredients without just putting 00:43:05.879 --> 00:43:09.139 your modeling hat on, but also in talking to your experimental colleagues, 00:43:09.279 --> 00:43:13.399 because things that you do have to make sense, see what could be responsible 00:43:13.399 --> 00:43:15.659 for the deviation between the immune experiment. 00:43:16.119 --> 00:43:19.379 This is the tedious cycle that we have to go through, whether in physics, 00:43:19.559 --> 00:43:21.179 biology, any discipline. 00:43:21.779 --> 00:43:26.419 That's correct. But now, so here we have this model. 00:43:27.239 --> 00:43:31.559 Right now, we're in a phase of the science that we actually try to understand 00:43:31.559 --> 00:43:35.179 how this system works at all, right? This is what you were clearing in your talk. 00:43:35.579 --> 00:43:40.699 There's a lot of aspects of this immune system and its responses that we just don't understand. 00:43:41.819 --> 00:43:45.759 But the next step would be to control this immune system. So imagine we move 00:43:45.759 --> 00:43:49.379 to this phase, now you have the model, we understand how it works, 00:43:49.439 --> 00:43:51.579 understanding would imply we can control it. 00:43:51.939 --> 00:43:56.719 So what kind of control would you envision we should pursue there? 00:43:57.719 --> 00:44:02.119 Well, I have a very simple but clear vision on that. 00:44:02.179 --> 00:44:05.279 I don't know whether it's realistic, but this is what I'm heading for. 00:44:05.279 --> 00:44:08.339 And that is in the immune system there's 00:44:08.339 --> 00:44:11.739 a very clear division of objects that 00:44:11.739 --> 00:44:14.999 immune cells can sense being categorized either 00:44:14.999 --> 00:44:18.019 as self or non-self as enemy or 00:44:18.019 --> 00:44:20.899 friend and we know one thing and 00:44:20.899 --> 00:44:24.439 that is the immune system is capable of changing 00:44:24.439 --> 00:44:29.059 its mind from one day to the next and this happens certain diseases when all 00:44:29.059 --> 00:44:33.299 of a sudden finds that the immune system starts attacking a particular type 00:44:33.299 --> 00:44:39.539 of cell cell and this is not desirable but we know that the immune system can 00:44:39.539 --> 00:44:44.239 do it what we do not know is what actually made that change how came that change about. 00:44:45.793 --> 00:44:50.493 So we first need to understand this. We need to understand what is happening 00:44:50.493 --> 00:44:53.873 when someone develops alopecia, where all of a sudden your hair fall out. 00:44:54.053 --> 00:44:58.593 Because it's a mechanism that is possible, but it's a mechanism that we would 00:44:58.593 --> 00:45:00.353 like to exploit and control. 00:45:00.733 --> 00:45:07.453 Because if I can manipulate this, if I can tell the immune system which objects 00:45:07.453 --> 00:45:10.673 are from today onwards to be seen as enemy, 00:45:10.813 --> 00:45:14.153 which until now was seen as friend, then we can 00:45:14.153 --> 00:45:16.913 redirect the immune system to attack anything we like 00:45:16.913 --> 00:45:19.733 and this is what i see as as really the 00:45:19.733 --> 00:45:23.193 future of medical applications of the immune system and 00:45:23.193 --> 00:45:26.693 this can be um especially relevant in 00:45:26.693 --> 00:45:30.073 cancer medicine where there 00:45:30.073 --> 00:45:32.873 are different levels of ambition one can have so 00:45:32.873 --> 00:45:35.633 one level of ambition i think is a bit too far for now and that is to 00:45:35.633 --> 00:45:38.493 train the immune system to see the difference between a healthy 00:45:38.493 --> 00:45:41.353 cell and a cancerous cell of a given organ 00:45:41.353 --> 00:45:44.593 but we don't need to do that we can also lower 00:45:44.593 --> 00:45:49.333 our ambitions and say look there are certain organs that a person can live without 00:45:49.333 --> 00:45:55.493 thyroid is one example if we could at least instruct the immune system to regard 00:45:55.493 --> 00:46:01.433 all thyroid cells as enemies then you could treat a patient first with surgery remove the thyroid, 00:46:01.673 --> 00:46:08.273 and then you switch your immune system to attacking all remaining residual thyroid cells. 00:46:09.478 --> 00:46:12.838 And the difference between that and the normal route, which is chemotherapy, 00:46:13.038 --> 00:46:21.118 is that if you do it right, you will not have any off-target pathologies that 00:46:21.118 --> 00:46:22.978 you create, any damage that you create. 00:46:23.278 --> 00:46:27.658 And it is an active agent. It is not a passive chemical. 00:46:28.518 --> 00:46:32.818 It is an active agent that will seek out anything that looks like a thyroid cell. 00:46:32.978 --> 00:46:38.058 So I can see this as the future. But isn't it, there's some molecular stuff 00:46:38.058 --> 00:46:41.118 going on there which is not captured in your model yet, presumably. 00:46:41.418 --> 00:46:45.178 No, these models are far from that stage. But I'm now looking ahead. 00:46:45.338 --> 00:46:50.518 This is one of the things that could become possible once we have an acceptable 00:46:50.518 --> 00:46:52.318 level of description of the immune system. 00:46:53.078 --> 00:46:56.578 Because I think we have to be extremely careful at the same time. 00:46:56.738 --> 00:47:00.278 Simply because the immune system is so powerful, and we've seen that in some 00:47:00.278 --> 00:47:03.758 trials that went dreadfully wrong. on, tinker with it the wrong way, 00:47:03.858 --> 00:47:05.298 you can do tremendous damage too. 00:47:05.558 --> 00:47:09.398 And it occurred to me that that's another use for your model because you can 00:47:09.398 --> 00:47:13.578 test interventions in the modeling environment before you try them. 00:47:13.678 --> 00:47:20.318 Because you explained how this particular disastrous trial started out with 00:47:20.318 --> 00:47:21.578 the successful animal models. 00:47:21.658 --> 00:47:24.598 So the animal models don't always predict what will happen in a human. 00:47:24.758 --> 00:47:28.958 But maybe a computational model will show you a variety of possible results 00:47:28.958 --> 00:47:34.778 and you you could then look out for bad outcomes that you might not have seen in the animal. 00:47:35.018 --> 00:47:39.678 Yeah, a colleague of mine from Brazil, from Berlin, once described it in the 00:47:39.678 --> 00:47:41.758 following way, which I think is a very good way of seeing it. 00:47:41.798 --> 00:47:45.538 If we build a new car and we crash test it, we don't run Mercedes against walls. 00:47:45.798 --> 00:47:48.698 We do it on a computer first. That's why we do the crash test. 00:47:49.878 --> 00:47:54.698 In medicine, we're not often doing that, sometimes because we don't have the 00:47:54.698 --> 00:47:57.718 computational ability or we don't have the knowledge to do it. 00:47:57.798 --> 00:47:58.998 But I think that is the future. 00:47:59.478 --> 00:48:02.938 Exactly as you say, you have a decent model and you have to be careful that 00:48:02.938 --> 00:48:06.818 you test your model against experimental data, but once you get to the stage 00:48:06.818 --> 00:48:09.698 where you trust your model to some extent, see if it exists. 00:48:10.820 --> 00:48:15.400 Then you can crash test ideas. And in the model, you can make those changes 00:48:15.400 --> 00:48:19.360 that you think on the basis of the theory would be the changes required, 00:48:20.080 --> 00:48:25.740 and then test whether it actually works like that or does have perhaps unforeseen 00:48:25.740 --> 00:48:27.720 consequences that are dangerous to the patient. 00:48:28.040 --> 00:48:32.580 I think it's more than crash testing as well. I mean, the driverless car research 00:48:32.580 --> 00:48:37.520 is now moving into simulation because they can't capture enough real-world data 00:48:37.520 --> 00:48:42.360 of unlikely things that will happen when you have millions of driverless cars. 00:48:42.440 --> 00:48:44.120 So you want to simulate those situations. 00:48:45.080 --> 00:48:49.960 But there are already attempts underway, as you also mentioned, 00:48:50.140 --> 00:48:56.520 also failed attempts, to try to really master and control the immune system 00:48:56.520 --> 00:48:58.860 for more personalized healthcare. Yes. 00:48:59.440 --> 00:49:05.160 So to what extent do you see these current steps as being consistent with the 00:49:05.160 --> 00:49:11.040 concept of the immune system that you are advancing with this model um i think 00:49:11.040 --> 00:49:14.120 it is going in the right direction in in early. 00:49:15.100 --> 00:49:19.200 Immunological therapy ideas very often 00:49:19.200 --> 00:49:24.160 what was done was non-cell specific interventions in the immune system which 00:49:24.160 --> 00:49:29.200 you can sort of see as similar to just rebooting it giving it a big kick and 00:49:29.200 --> 00:49:33.760 then hope that it settles into a new state which is doing what it was designed 00:49:33.760 --> 00:49:36.480 to do and more recently differently, 00:49:36.480 --> 00:49:39.100 especially with this example of these CAR T cells, 00:49:39.460 --> 00:49:46.480 the research has moved away in the direction of making genetic modifications 00:49:46.480 --> 00:49:53.200 that are not generic, but they are tailored to the specific needs of an individual patient. 00:49:53.520 --> 00:49:59.640 And in the case of CAR T cells, they create these receptors to be specifically 00:49:59.640 --> 00:50:04.060 tailored to recognizing and docking to the cancer cell. 00:50:05.120 --> 00:50:08.600 Now that is moving already in the right direction. It's the only thing is it's 00:50:08.600 --> 00:50:12.500 fairly limited because this only is done for cytotoxic T cells. 00:50:12.700 --> 00:50:17.200 So this doesn't exploit at all the interaction, the network concept, 00:50:17.300 --> 00:50:19.860 the interaction between the T clones and the B clones. 00:50:21.540 --> 00:50:26.140 And in that sense, it is still very different. It's not this idea that I have 00:50:26.140 --> 00:50:28.280 of controlling the self non-self boundary. 00:50:28.440 --> 00:50:31.500 That's not what's happening, but at least it 00:50:31.500 --> 00:50:35.520 is moving in the the right direction it is trying to make non-generic changes 00:50:35.520 --> 00:50:40.440 but it's trying to make changes to components that are really tailored to the 00:50:40.440 --> 00:50:45.200 specific needs of an individual and they are spectacular some of these results 00:50:45.200 --> 00:50:49.600 it isn't doing everything we would want these things to do the CAR T cells don't 00:50:49.600 --> 00:50:50.980 work for solid tumors at the moment. 00:50:52.590 --> 00:50:55.410 There's a lot of research going on in seeing what is the 00:50:55.410 --> 00:50:59.030 reason why it doesn't work there what can we change and there 00:50:59.030 --> 00:51:02.010 have been already several generations compared to the first generation 00:51:02.010 --> 00:51:04.650 of these methods they've come a long 00:51:04.650 --> 00:51:09.910 way and that's why they've now been uh approved by the various authorities right 00:51:09.910 --> 00:51:14.690 but why why would this work on by manipulating the t cells and not the b cells 00:51:14.690 --> 00:51:20.250 because you could imagine that the b cells that was labeling um whatever object 00:51:20.250 --> 00:51:22.690 you would like to classify as non-self. 00:51:23.350 --> 00:51:26.670 So can you explain from the perspective of your model why it has been effective 00:51:26.670 --> 00:51:28.330 on the T cells and not the B cells? 00:51:28.630 --> 00:51:31.550 That's a very tricky question because I don't really know the answer to that. 00:51:31.650 --> 00:51:34.950 And I can only speculate as to why that may be. 00:51:35.150 --> 00:51:37.230 But given the model, you must have an intuition, no? 00:51:38.830 --> 00:51:43.930 Well, let's approach it from the other side. Would it be a viable alternative to do that? 00:51:44.630 --> 00:51:49.570 I think it would be. But I need to look very carefully at the differences, 00:51:49.570 --> 00:51:55.410 is if any between the response of the immune system and the way it does its business in, 00:51:56.370 --> 00:52:02.290 attacking foreign antigen compared to how it deals with deranged own cells and 00:52:02.290 --> 00:52:07.010 i would not be surprised if let's say the players responsible for the immune 00:52:07.010 --> 00:52:08.410 response would not be the same, 00:52:09.470 --> 00:52:14.750 because b cells are being trained very specifically in the germinal center to 00:52:14.750 --> 00:52:20.470 look away from the cell cells, to really look only at foreign antigen. 00:52:21.230 --> 00:52:26.590 So it may well be that the way the immune system handles cancerous cells is 00:52:26.590 --> 00:52:32.150 different, distinct from the way it is most effective in handling foreign antigen. 00:52:33.731 --> 00:52:37.631 There must be studies looking at that, but I'm not aware of them at the moment. 00:52:37.771 --> 00:52:38.771 But it is a very good point. 00:52:43.071 --> 00:52:48.831 So you move from surgical physics to neural networks to the immune system. 00:52:49.471 --> 00:52:54.211 Do you in any way miss the study of neural networks and the brain, 00:52:54.351 --> 00:52:58.471 or not that much? Do you find this more challenging? 00:53:01.231 --> 00:53:05.111 That's an interesting question again. and I've had 00:53:05.111 --> 00:53:08.971 a really good time studying neural networks but I 00:53:08.971 --> 00:53:11.931 also find very often that when I've worked 00:53:11.931 --> 00:53:14.811 on something for 10 years or more I tend 00:53:14.811 --> 00:53:18.871 to look for something different and it 00:53:18.871 --> 00:53:22.111 can be that different in terms of the modeling 00:53:22.111 --> 00:53:24.911 and the analysis can be found within 00:53:24.911 --> 00:53:28.391 the same domain main or outside and 00:53:28.391 --> 00:53:31.811 and what i've seen at the time is that i 00:53:31.811 --> 00:53:35.231 found all kinds of new things to work on which happened not to be neural networks 00:53:35.231 --> 00:53:42.831 but now just by accident i'm moving back and i'm touching that field again simply 00:53:42.831 --> 00:53:49.731 because the technology leads me there um i've always enjoyed I thought it was 00:53:49.731 --> 00:53:51.431 a fascinating area to work in, 00:53:51.471 --> 00:54:01.731 but at some point I got the impression that in this community it became an inappropriate 00:54:01.731 --> 00:54:05.011 norm that one would have to know everything about biology, 00:54:05.311 --> 00:54:08.951 everything about the computer science aspects, everything about the theory, 00:54:09.971 --> 00:54:14.151 in order to be effective. I believe that's a fallacy. 00:54:14.851 --> 00:54:19.311 And I think it has done some damage to that field and it has sort of pushed communities apart. 00:54:21.461 --> 00:54:25.441 I find these things very interesting because I've now, about 10 years ago, 00:54:25.521 --> 00:54:30.561 moved into what I call quantitative medicine, did a lot of work on protein interaction 00:54:30.561 --> 00:54:32.101 networks, gene regulation. 00:54:33.761 --> 00:54:37.881 And I've asked myself, how can we avoid that in the future? 00:54:39.001 --> 00:54:44.581 So what we've done is we've set up a mechanism where young people are given 00:54:44.581 --> 00:54:49.241 a program, a training program, in which they are being taught by people from 00:54:49.241 --> 00:54:52.601 different disciplines, at a level that they can understand. 00:54:52.901 --> 00:54:59.501 And we try to educate them to work differently, to not believe that in working 00:54:59.501 --> 00:55:02.741 in an interdisciplinary team, one has to know everything about all the disciplines 00:55:02.741 --> 00:55:04.981 because one ends up being mediocre at everything. 00:55:05.601 --> 00:55:10.661 But they have to identify what they are in terms of their primary discipline 00:55:10.661 --> 00:55:12.761 and make sure that they are really on the ball in that. 00:55:12.961 --> 00:55:18.101 But they need to speak the jargon of the people at the other side of the various 00:55:18.101 --> 00:55:20.961 fences, experiences, and they need to know what keeps them awake at night. 00:55:21.401 --> 00:55:24.921 And that's the only crucial thing. But they should not make the mistake to think 00:55:24.921 --> 00:55:27.521 that they have to be experts at three different disciplines. 00:55:27.981 --> 00:55:32.281 It's not realistic. It's counterproductive. I think at the time, 00:55:32.341 --> 00:55:35.641 but maybe I'm wrong because you know that community better than I do, 00:55:36.661 --> 00:55:40.421 there was a bit of that happening in between the 1990s and 2000s. 00:55:42.221 --> 00:55:50.041 It's still like that. Yeah. Yeah, yeah. I mean, multidisciplinarity is clearly 00:55:50.041 --> 00:55:52.221 still a challenge also in the field of neuroscience, 00:55:52.521 --> 00:55:58.241 where there's from the side of the people doing the experimental work, 00:55:58.481 --> 00:56:01.421 which is extremely intense. 00:56:02.061 --> 00:56:06.841 Of course, they would feel like, well, these guys haven't made the miles to 00:56:06.841 --> 00:56:09.261 really appreciate the complexities of what we're doing here. 00:56:09.361 --> 00:56:13.021 They will not really appreciate the meaning of this kind of data. 00:56:13.821 --> 00:56:18.021 And the modelers or the theoreticians would often have the feeling like, 00:56:18.061 --> 00:56:21.261 well, we're not being listened to, right? So, there's this massive disconnect. 00:56:21.521 --> 00:56:26.981 And that's also why I think in neuroscience, theory overall has made less of 00:56:26.981 --> 00:56:28.601 an impact than you might have wished. 00:56:30.041 --> 00:56:38.181 But now, so, okay, given this long trajectory you've made, what should be Ton's 00:56:38.181 --> 00:56:44.541 law if we want to understand the immune system, the brain, living systems? So, what's Ton's law? 00:56:45.441 --> 00:56:48.781 Well, I don't think I have one law. I've learned a couple of things, 00:56:48.861 --> 00:56:52.481 and I can go through them if you want. The first is interdisciplinary interactions. 00:56:54.004 --> 00:56:59.864 What i just explained to you and i think where we should move to is a is a way of working based on, 00:57:00.424 --> 00:57:04.884 respect of expertise on the other side right if i work with an experimentalist 00:57:04.884 --> 00:57:10.744 a bioinformatician i have to trust them to know what they do and i i leave them 00:57:10.744 --> 00:57:12.804 to it what i do need to know is, 00:57:13.464 --> 00:57:16.284 what is it they really find interesting but what i 00:57:16.284 --> 00:57:19.184 should not try to do is to convince them what they should find interesting 00:57:19.184 --> 00:57:21.984 is what i find interesting yeah so these 00:57:21.984 --> 00:57:25.664 are the types of of psychological things i've learned that means empathy not 00:57:25.664 --> 00:57:28.444 not that empathy i just need to know what what excites you 00:57:28.444 --> 00:57:31.244 paul i need to know what what is it that makes you really 00:57:31.244 --> 00:57:33.964 happy you need to know what makes me happy and we have to accept that these are 00:57:33.964 --> 00:57:38.184 different things but they could all be found in the same project but they can 00:57:38.184 --> 00:57:41.944 all be found in the same project and that's how we can work together best and 00:57:41.944 --> 00:57:46.724 i don't and i should not try to do someone else's work or even attempt to know 00:57:46.724 --> 00:57:51.664 what they know this is completely counterproductive so that's that's that's one of the first things. 00:57:54.144 --> 00:58:01.144 Generally, I think we live in a bit of a strange time in terms of how science 00:58:01.144 --> 00:58:05.304 is organized generally, and I think there will come a point where there will be major revisions. 00:58:05.984 --> 00:58:10.264 One of the crucial lessons is never sacrifice quality. Everything else you can 00:58:10.264 --> 00:58:12.564 sacrifice, but never sacrifice the quality of what you do. 00:58:12.644 --> 00:58:17.804 Be your own worst referee yeah because fashions 00:58:17.804 --> 00:58:22.444 will come and go reputations will come and go but you want your name to be a 00:58:22.444 --> 00:58:28.684 sign of quality the only way to protect that is to never sacrifice quality and 00:58:28.684 --> 00:58:33.844 if you then perhaps are six months later in submission of a paper and it doesn't 00:58:33.844 --> 00:58:35.024 end up in nature who Who cares? 00:58:37.216 --> 00:58:39.836 But maybe in 15 years from now, they will still read that paper. 00:58:40.096 --> 00:58:44.856 And all the nature papers will have been seen and forgotten about in six months. 00:58:45.496 --> 00:58:48.576 Quality is everything in science. That should never be sacrificed. 00:58:49.276 --> 00:58:52.856 Okay. Another loss you want to add to that or two is enough for now? 00:58:53.156 --> 00:58:56.096 No, I drink lots of coffee, Paul. Okay, good. 00:58:57.076 --> 00:59:02.716 So to finish up, so we're going to send Tony from Sheffield to London five years from now. 00:59:03.556 --> 00:59:08.876 Okay. Okay, given the current political situation, let's just assume it's London, 00:59:09.016 --> 00:59:09.716 it might be somewhere else. 00:59:10.116 --> 00:59:12.636 You never know, with Brexit hanging in the air. 00:59:14.116 --> 00:59:20.016 But okay, there comes Tony, he visits you at King's College to check whether 00:59:20.016 --> 00:59:23.716 a specific hypothesis was falsified or verified. 00:59:24.396 --> 00:59:27.296 And that is the hypothesis that you're going to specify today. 00:59:27.416 --> 00:59:32.856 So what's the one hypothesis that you really would like to see tested in this five-year frame? 00:59:32.856 --> 00:59:40.716 That it is possible to instruct the immune system to change which object it 00:59:40.716 --> 00:59:42.656 sees as self and it sees as non-self. 00:59:42.956 --> 00:59:46.856 I think that would be a big, big step forward if we can achieve that and actually 00:59:46.856 --> 00:59:49.996 have experimental validation for it. 00:59:50.576 --> 00:59:53.476 Great. Tom Cullen, thank you very much for this conversation. 00:59:53.676 --> 00:59:54.836 Thank you. Thank you very much. 00:59:57.616 --> 01:00:03.116 The CSN Podcast was produced by the Convergent Science Network of Biometrics 01:00:03.116 --> 01:00:09.816 and Biohybrid Systems, a project funded by the European Sevens Research Framework Programme. 01:00:11.016 --> 01:00:16.336 For more interviews, recorded lectures or upcoming conferences in the field 01:00:16.336 --> 01:00:22.576 of biometrics and biohybrid systems, go to csnnetwork.eu. 01:00:22.800 --> 01:00:30.480 Music.