Transcript

Building a model brain: Richard Walker at TEDxMontpellier

[Music] [Applause] thank you very much indeed it's a great pleasure to be here in moner today in your wonderful City uh so I'm the spokesperson for the human brain project which as Mel has just said Has just uh won itself a bit more than a billion euros of funding from the EU and you might not be surprised to know that the theme of the human brain project next slide please the theme of the human brain project is the human brain which some people think and many people have said is the most complex organ on Earth it's this organ which allows us to organize meetings like this it's this organ which creates music it's this organ which speaks which writes poetry and it's an organ which next slide has something like 80 billion neurons and it's 80 billion individual cells which are connected by something like 80 trillion uh 8 trillion connections or synapses but there's something which is particularly miraculous about all that even though this you have this Forest you can the genome which contains the blueprint of the human brain can be written down in about one and a half gigabytes one and a half gigabytes for a computer person is not very much if we just go on and look at this machine here this machine which is a very oldfashioned one CU I buy cheap ones uh has 32 chica bits in it so in just 5% of the space here I can put the whole of the blueprint actually for the whole human body but for the whole human brain and that is a mystery and that leads to two questions there's a scientific question which is how on Earth do I go from this very relatively short compact description to this immen ly complex organ which writes poetry that's a scientific question that's fascinated me my whole life and then was a practical question how can I find out now one of the greatest scientists in the greatest physicists in the last uh 100 years was U Richard feineman and he said if you want to understand a complicated system you should build it so that's what we do we build brains what that actually means is we build simulations of the brain models of the brain in huge supercomputers and you start getting out things like this maybe I can bang something and it will come uh you get these wonderful simulations now the theme of today is actually medicine so you have a right to ask what's all this got to do with medicine well if your iPhone breaks down you if you want to mend it it's not good enough to just know how to use an app you have to open it up and look at all the pieces inside and see which piece isn't working you have to actually understand the internal mechanics of your iPhone and we think simulation and modeling is the way we should do that now there's a lot of objection to this Vision some people say but look you've got these 80 billion 80 billion neurons and you've got 80 trillion copses you know measuring them is difficult we know that if you measure all of them you'll just never get there it's going to take a centuries to do it so I say you're you're you're trying to do something completely impossible but I'd like to remind you that the description all of of all that is this one and a half gigabytes which fits 20 times inside this little phone and that suggests that there's some underlying Simplicity there there's some basic principles and it's the existence of those principles that mean that a project like ours is feasible so what do weang what do we actually do yeah uh what do we do we sort of fill in a jigsaw puzzle we get data about the brain from all over the world we get data about brains of different species different parts of the brain we get different levels of detail some people tell us about the molecules other people tell us about the circuits other people tell us about the forms of the neurons and we put it all together in a jigsaw puzzle and as we do so slowly these principles of Simplicity which make the modeling possible begin to fall out and I'd like to just tell you a little story about just one bit of this which we're actually doing ourselves uh it's been known for a long time that neurons come in different shapes typ shapes and sizes this is not something we discovered ourselves but you have pyramidal neurons you have basket cells you have Marty nauy cells and you can carry on naming them for a long long time so we start off with a taxonomy of neural types A classification then we start counting them in our particular research in this particular bit of the story we look at the cortical Colum of a WRA that's a very small part of the cortex of a WRA which is divided into six layers and we count how many cells of different types are in different parts that's difficult but it's definitely doable so that's stage one uh stage two we get actually get Neons we stain them we color them we put them under a microscope and very painstakingly we make digital models of the shape of the neuron the three-dimensional shape we make the threedimensional shape just like the figures you see a video games you can turn it around turn it upside down shake it around do anything you like with it and we uh measure its electrical Behavior using this wonderful equipment here which is called a 12 patch machine you don't have to remember that which actually measures the electrical behavior of the cells now once we've got all that a little miracle happens because what can we do we can create a model you we take neurons in the right proportions so so many paramal cells so many of this size so many of that the right proportions for that bit of the brain and we put the digital models inside our virtual space and lo and behold we find that the fibers of these models touch and we know when when they touch that can form an electrical circuit it so instead of measuring every individual sinapse which we agree with everybody else is completely we can't be done what we've actually measured is the shape of a finite number of cells we've done some counting and we have the circuitry falling out on its own it's not quite perfect but it's very very good indeed and we know that actually in real life that circuitry that is fine-tuned by the person's experience but I'll come to that later and once we've got a model of the circuit bang we can move it onto a supercomputer and we can actually simulate how that circuitry works so we can inject a simulated current into the circuit and then some cells will start firing their signals will travel down the fibers go to other cells they'll start firing too and so on producing these wonderful pictures we have here and we can take that model we can examine it statistically we can we can compare it to the data we have from real circuits we find out what's wrong because it's not perfect we correct it and we slowly get better and better now one of the things which makes our work possible is because the brain has lots of repetition you don't actually need to do to measure every neuron because they're very similar they're even similar between different species or different parts of the brain and we don't have to measure every single individual iCal column CU they too are very similar certainly they're not identical but we can take a basic model and then we can tweak it we can make it more perfect and so we can build bigger models much faster just like you know moving from a single transistor which they had in 1957 to this thing which has many millions it didn't take millions of years because they were repeating themselves many many times and we can take information from other groups and we can look at the long range connections between these different areas of the brain and start those putting those in the model as well until we get to a whole brain this is not a human brain this is a rat brain that would already be a huge achievement we're not there yet but we can gradually build up a complete detailed working model of the healthy human brain now that's the healthy brain but the theme for today is medicine and in particular it's diseases of the brain now there was a report uh last in 2011 from the European Brain Council which estimated that the total cost of brain disease for the European economy is something of the order of 800 billion euros every year and we're talking about diseases like depression uh like schizophrenia like autism uh like anxiety like the degenerative diseases like Alzheimer's and these diseases we're in a bad way today because we have great difficulty in diagnosing them we don't have a lot of knowledge about their causes we can manage them we've much improved the standard of living of people who have these diseases but we can't cure them they G on forever and actual Research into these diseases is being reduced because the drug companies have had a huge number of failures they're very expensive failures and they think it's better to to invest in things where they have a higher chance of success so we want to use our models to work on these diseases how well when you have a disease of the brain it's because something's gone wrong with the brain that doesn't mean necessarily that the cause of the disease is a gene or chemistry it doesn't have to be that way maybe it's something in in the environment but even uh when it's the environment which is causing your disease this affects the brain it changes it physically if you have a kid who's been grossly abused you're going to find a lot of changes in that kid's brain you might find a reduced hippocampus you'll find a thing called the HPA axis is going to work differently from usual so you have a physical disease but when a doctor sees a physical brain disease he sees lots of things he sees the original C cause of the disease he seizes things which have been caused by the disease he also sees changes which have been caused by the drugs used to treat the disease and it's extremely hard to untangle all this now this is where modeling comes in what do we want to do just like we collect data about the healthy brain from lots of different sources we want to collect data about the diseased brain hospitals today if you go and you have a brain scan the neurologist will look at it for a few minutes and make a diagnosis and it will go into an archive there it will remain and probably no one will ever use it again we'll ever see it again we would like to take this data anonymize it so you don't have people looking at it for reasons they shouldn't be looking it at and um make it available to medical researchers so the first job is to look find groups of patients who have similar conditions defined in biologic iCal terms and extract what we call biological signatures of disease now biological signature is a set of changes in the brain which differentiates that patient compared to a patient with a different disease or compared to a healthy patient then we can take those and we can start playing with our model we can take one aspect of that signature and tweak the model to reflect it so say we think that the neurons are hyperactive we can tweak our model to make the model neurons neurons hyperactive as well we see what happens maybe it doesn't produce anything nothing happens so we know that was some secondary effect which doesn't matter so much maybe it produces the whole Cascade of effects we're getting in the disease then we found a cause and once we have a model of the disease then we can also model possible treatments now and and obviously the goal is to get better treatments now we don't want to promise miracles uh this is going to be a long slow process and nothing nothing we do gets rid of traditional medical proes even if we find interesting things or if people who use our platform find interesting things you'll still have to test them in animals you'll still have to do clinical trials it will still be a very long slow process but our goal is to accelerate the search for new treatments to make it better and it's a very simple calculation you can do if we could reduce the cost of brain disease just by 1% and that's not being too ambitious that's 8 billion EUR a year in Europe alone and that would pay for our whole 10year project eight times over so we think we're on to something now that is actually my most important message but before I end before I end uh I want to address some other questions things which people always ask us and which I think we shouldn't try and escape from so people say you're trying to build a model of brain on a huge supercomputer and they say are you going to make an artificial brain is it going to take over the world is it going to be more intelligent than us is is it dangerous you know is it going to be an alien uh well that's a very very straightforward scientific answer to that the answer is no and it's not just because we're extremely nice people though of course we are uh it's for a very good scientific reason I said the brain was plastic now the brain changes in every moment of the life of a child the life of an adolescent the life of an adult of an old person every time you hear a sentence every time you eat a food you like every time you listen to some wonderful music music your brain changes some synapses get stronger some get weaker some disappear some new ones form and it is this process of continuous change in the brain which makes us what us what we are as human beings now we don't actually model that process we don't do it Part largely because we don't have the knowledge to do it we have to know our own limitations partly we don't have a computational power to do it either so this is something way way out in the future for the moment it's science fiction we can't model development and for that reason we are not going to produce artificial brains which take over the world which means of course that my original Dream understanding how to get the one and a half gigabytes to write poetry is not completely resolved we're not I'm not going to see that I don't think but you know I'm a scientific romantic I got interested in science when I was about 8 years old and I saw I listened on a very ly wooden radio to Alan sheeper being the first American in space a few years later I was uh I stayed up with my parents very late at 4:00 a.m. in the morning to watch the first man walk on the moon and that excited me and I'm excited by understanding this problem here and I think we are going to do some get some understanding of it so for me now I'm personal more than the project we do have a moonshot we want to at least be able to reproduce simple bits of human brain function if we could reproduce just the first acts of a baby when it starts to learn to talk that would be resolving a scientific problem which um has been open for 2,000 years how do human beings think no one knows we want to give put new light on that question with lots of medical results coming coming out from what we're trying to do so that's our moonshot that's why we've asked for the money and if we can get there I think it will be a small step for our project but it really will be a very big step for human knowledge thank you