Psychiatry is changing and will be unrecognizable in the next 10-20 years, given our new understanding about the role of brain circuits in the generation of emotions and behavior. This week we talk to Professor Karl Deisseroth, D.H. Chen Professor of Bioengineering and of Psychiatry and Behavioral Sciences at Stanford University and the Howard Hughes Medical Institute.
Psychiatry is changing and will be unrecognizable in the next 10-20 years, given our new understanding about the role of brain circuits in the generation of emotions and behavior.
This week we talk to Professor Karl Deisseroth, D.H. Chen Professor of Bioengineering and of Psychiatry and Behavioral Sciences at Stanford University and the Howard Hughes Medical Institute. We discuss his work on optogenetics and the insights it has given into the workings of the human brain.
Professor Deisseroth also explains the potential for the future of psychiatry: including deep brain stimulation, transcranial magnetic stimulation, and the use of our smartphone and digital data, known as our “digital exhaust”. Will these methods become a regular part of psychiatric practice in the future?
This season we'll dive deep into the languages of life through explorations of the "dark genome", genome editing, protein folding, the future of aging, and more.
Hosted by Anthony Philippakis (Venture Partner at GV) and Alex Wiltschko (Staff Research Scientist with Google AI), Theory and Practice opens the doors to the cutting edge of biology and computer science through conversations with leaders in the field.
Hello, you're listening to Theory & Practice. I'm Anthony Philippakis.
And I'm Alex Wiltschko.
Today we are going to talk about the future of psychiatry. Our understanding of the way emotions and mental distress are generated has changed fundamentally over the years. In the 17th century, René Descartes espoused the view that the mind existed beyond the body. This dualism separated psychiatry from other forms of medicine, whereas blood tests, imaging and physical examination are used to diagnose many diseases. The diagnosis of mental illness has traditionally been based on consultation alone.
This reliance on the spoken word makes the task of identifying different psychiatric illnesses, sometimes messy and unreliable affair. As a result, patients don't always get the best treatment.
During the 1960s Descartes’ theory of dualism began to be challenged. Physical treatments such as antipsychotics were increasingly used to treat mental illness. The creation of new antidepressants gave credence to the idea that some mental illnesses were due to chemical imbalances. To be fair, that idea is now being challenged by a greater understanding of the role of neural circuits.
Neural circuits are groups of brain cells that are linked together by synapses to form networks in the brain that support or discharge a particular function. In a computer, you have hardware, like a CPU and a hard disk, and you've got software which defines what the computer should be doing. But in the brain, the hardware is the software, the software is the hardware. So what's happening in mental illness, when that system isn't working well, in neural circuit that supports a particular function, like enabling you to respond with fear to a dangerous situation can sometimes go wrong, can be overactive, or even underactive, so for instance, someone with PTSD has neural circuits that might cause them to replay images over and over again, from a traumatic event. Our guest today, but the development of one of the most important tools for understanding these neural circuits optogenetics a way of using light to turn on and off neurons by embedding light sensitive bacterial DNA and proteins into a virus that is taken into brain cells and makes them sensitive to light. Professor Karl Deisseroth is the D.H. Chen Professor of Bioengineering and of Psychiatry and Behavioral Sciences at Stanford University, and the Howard Hughes Medical Institute. He's recently written a book called Projections: A Story of Human Emotions. The title of his book is a play on the word projections since in psychiatry, it means, traits in your own character - usually negative ones - that you portray as belonging to someone else. And in neuroscience, it means nerve cells that project from one area of the brain to another, can both relate to and support the generation of emotions.
Today, we'll explore how tools like optogenetics have helped us understand where emotions are produced in the brain. We'll also discuss where other developments in the field of psychiatry fit into this central understanding of brain circuits. We'll talk about developments such as implantable deep brain stimulation, new uses of psychedelics and the treatment of depression, and the emerging science of digital psychiatry. What does this all mean for psychiatry in the next decade? We will also cover that.
Karl, welcome to Theory & Practice.
Thank you; looking forward to talking.
So to kick us off, let's start with a very basic question about your research. What exactly is optogenetics? How did you find out about it?
So optogenetics is a technique. It's a method that we use in biology, it uses light to make things happen within animals, within specific cells of living animals. And this is something we developed in my lab beginning about 17 years ago. Now, it's the opposite of how you normally think about the use of light. You know, when you think about light, you think about collecting information, you know, we think about looking through a microscope collecting data. With optogenetics, it's the opposite of that we're using light to make things happen - to put them into the system and make cells turn on or off to make specific kinds of neurons, or even a specific, individually specified neuron, fire with a specified pattern, all in the context of free behavior of complex animals, including mammals. And so that's valuable because it tells us what actually matters: what's causal, what can and what does matter for all the complex, wonderful things in the brain, sensation, and cognition and action.
Over the last 17 years, you've leveraged this tool to make a series of seminal discoveries. Maybe you can outline a few of the ones you're most proud of.
There are so many exciting things to choose from. So it's a good question, but I'll tell you. Because I'm a psychiatrist - what I'll talk about is one particular finding that touches on anxiety. And, I see patients still one day a week. But most of the time I'm in the lab,thinking about these clinical issues that my patients are experiencing. And anxiety is one of them - it's actually quite hard to treat - it's very common, a little bit of it is good; too much of it is bad. And it can become incredibly debilitating and harmful. But it's a complex state, it's got different features to it, you've got the physiological changes, the heart rate goes faster, the breathing, respiratory rate becomes faster. And then you've got behavioral changes, you've got avoidance of risky situations, even if there's not an immediate threat. And this is one of the most debilitating things, and people with anxiety - many of them can't leave their house. So avoidance of perceived risk - so, behavior. And then finally, it's got a negative quality to it - it feels bad. Okay, this sounds obvious, but actually, it's quite an important and subtle point. Why should nature make us feel bad? If it's already making us avoid the risky situations, why do we have to also have that negative internal state as well? So it's all these different features, and they come on together, and they go away together? And how is this created? How was it coordinated? How was it made into a coherent state? This is what optogenetics helped us sort out. You know, we had developed the core technology of optogenetics, over about five years from 2004, to about 2009. And then we really started applying it to questions about how our brain states and behavioral states created and formed and then broken down. And in 2013, we published a paper on anxiety. And we showed that there's a region in the brain - in the mammalian brain - called the bed nucleus of the stria terminalis (BNST). And it's a little extended part of the amygdala - among other things, it's involved in anxiety and fear responses. But we found that the state of anxiety can be put together, assembled and disassembled by different projections coming out from the BNST and going to different parts of the brain. We found one kind of cell that sends one connection to an area of the brain called the lateral hypothalamus. And that controls the behavioral avoidance feature of anxiety. Another projection goes from the BNST to a spot that regulates breathing deep in the midbrain. And this is the parabrachial nucleus, this controls the respiratory rate changes. And yet a third projection goes to a part of the brain that seems to set this negative internal valence sort of sign of the state - the negativity of it, and that's in the ventral tegmental area. And this was a finding that was accessible. Because of optogenetics, we were able to turn on or off the individual projections going from the central control region to these different outlying regions and see they very cleanly recruited one only, not the other two of these features. And so we can see how the state was assembled. It's tractable, it's rigorously tractable, you can define it in a material physical way. And after that paper came out, I got contacted from patients around the world, people suffering with anxiety - they understood that it wasn't yet a treatment, but they were so grateful and hopeful that this reduction of this complex, messy state to its elemental components - was possible.
You know, it's so fascinating. I mean, it's really where mind meets brain. Now, you mentioned a lot of patients reaching out to and that they were excited about what it could mean for therapies. I mentioned, you might also get the reverse, however: which is the people who kind of find it difficult to believe that something as complex as emotion can be reduced to a mechanistic neural circuit? How do you respond to them?
Well, I am a great admirer of the human brain and its incredible richness. And I'm the first to state that there's a mystery that we have not come close to understanding. That is one of the exciting things about being a neuroscientist; we have a long way to go. So that would be the first thing to say. The second thing to say would be that it's just enthralling to see what we can figure out, understanding that we haven't figured everything out - not nearly. But so many labs around the world are using these methods. Catherine Dulac’s group at Harvard used the very same method I just described to break down parenting - this complex, quintessential mammalian state of parenting. And she was able to show with her team that one projection controls and is able to mediate, the going out and retrieving the young that have strayed too far from the nest - but doesn't affect the grooming the physical care of the young - and another projection mediates the physical care, but doesn't mediate the retrieval of the young. And as a parent, I know those two parts are very important and yet they've always seemed wrapped up together, but to see how they're separable is just a fascinating thing. I would say to people who say, you know, there's incredible complexity - I would say absolutely yes, but let's see how far we can go with this causal material, elements of approach. It's amazing what we can discover.
So one thing that that I'm, I'm struck with listening to you is this returning to the concept of taking something which is holistic, a human experience, and then finding that it's made of pieces. I'm curious if your perspective about yourself, or about the patients that you've treated has changed as you've kept finding this atomic nature of emotions, of behaviors, again, and again. How has that changed your perspective on what you do and how you interact?
You know, it's helped me realize what I take for granted. And so much of science is like this. So much of the world that seems to fit our intuition. We don't question it. We don't take it for granted. And then when we start to ask, Well, why really, is it that way? And how could it not be that way? That's when really deep insights appear. That happens in physics, happens in cosmology, happens in chemistry, and happens in biology. And as one example, something that you take for granted is the nature of the self. You asked about my own introspection? Well, I've always just thought - myself as myself -I am who I am. No need to question that further. But it turns out that the self can be disassembled. And not only that - in a tractable, rigorous, quantitative way. And this is something that optogenetics again, has helped us with, but the experience of the disassembly of the self is much more universal. People who have been through severe stress or trauma have experienced this phenomenon of dissociation, where they feel separate from their body, their sense of self and their body becomes separate. More than 70% of people who go through severe stress or trauma will experience this phenomenon. So this is a part of how the mammalian brain is designed to be able to separate the self. That's something most of us don't question in the course of our life, that the different parts that we consider to be us might be disassembled. And so it's a fascinating thing to consider, and what I have found in introspection about myself, what we've learned is: take nothing for granted. Look at how things are and ask why. And it's been incredibly exciting and productive and illuminating to do that.
So if I can just jump off of that…the invention of optogenetics itself has some of those characteristics of not taking things for granted. How did you come across the tool of optogenetics in the pieces that it was kind of existing in, in life? And how did you assemble that together?
Well, as I said, it took about five years to put the pieces together. And one of the reasons was that each of the individual parts hadn't been pushed quite far enough to be useful together. And that had created a big energy barrier in the field. Quite ironically, the core wetware, which is something called a microbial opsins gene - that is made by microbes. This had been known for decades, this was sitting there not even hidden in nature, but part of the textbook training of chemists, and I learned about the microbial opsins in Lubert Stryer’s beautiful biochemistry textbook, that the microbial opsins had been identified in 1971. People knew that there were these light activated proteins that receive a photon of light and move charged particles across the membrane of cells. So they're beautiful little molecular machines, they receive a photon of light, and they move ions across the membrane of cells, which is electrical current. And that is something that microbes do for their own reasons. They do that to make energy; they do that - follow light - for whatever they need to do for survival. And there's a huge family of these microbial opsins, studied in detail, hundreds of papers over the years, and beautiful work in the biochemical realm. But the currents that they made were very small. And these were from microbes that are very distant across the tree of life from mammals. And the motivation to try using these as a control tool was greatly tempered by a number of other factors. The currents are sort of small; you'd have to jam a lot of these into the cells to make enough current to be able to turn a cell on or off. That's probably particularly for this foreign protein, from a microbe, but that's probably not going to be safe, that's probably going to damage the cell. How are you going to make it specific also? This is the whole essence of optogenetics. We don't want to turn all the cells on. We just want to turn on the ones we're interested in at that particular moment. So you've got to use genetic targeting strategies to get the microbial opsins just into the cells that you want. And at the time that I started this work in my lab, there was not a real robust or reliable, versatile way of doing that of getting huge quantities of a highly expressed protein safely expressed just in the cells of interest. And then finally, the whole light delivery - how do you get the light into the right spot of the brain targeted to the right cells? At the time, LEDs weren't strong enough to generate the right power levels, and so there were a whole host of different things: the targeting, the light delivery, and the wetware. All had to be worked on. But none of them had advanced enough by themselves. And so that's why it took a number of years to put all the pieces together. And so it's a very perceptive question because optogenetics now looks like a coherent whole. But it had to be assembled from its parts, which were not enabled for use with each other when we started.
I remember that moment in time; I was just starting my training in Systems Neuroscience. We had all the electrophysiology equipment you could possibly want, and then none of the other stuff that we needed. And it was just so clear at the time after that kind of remote controlled mouse experiment that this was just a future. I mean, we, you know, we were doing stimulation and in basal ganglia, and we kind of knew some of these cell types, and kind of knew their genetic identity, but it changed the game from looking at the shapes of the squiggles of these neurons and giving them names to looking at their genetic identity and controlling them.
Yeah. One thing I want to double click on is your former student, now Professor Viviana Gredinaru. You mentioned before your work on anxiety, finding all these different projections out, and each projection carrying an aspect of fragment a shard of anxiety. But Viviana's work is interesting, in that, she was studying deep brain stimulation, and the frequencies at which these stimulations need to occur in order for them to have clinical effect. And I thought that was just an interesting, complementary aspect that, not only are the projections important, but how they're stimulated over time is important. So I just kind of want you to reflect on that a little bit with us.
Yeah, that is a critical aspect of optogenetics - the speed, the temporal precision. Scientists and biologists have always had ways to target cells generally, even in a cell type specific way you can put in ion channels that turn up or down stably, the activity of cells - that's been done for a long time, you can of course, kill them even in a genetically targeted way. But what optogenetics brought was the ability to interact with the brain to exchange information at the relevant speed at the frequencies and with the timing, that actually are extremely important for brain function. And that works because these microbial opsins encode proteins - they are very fast, and light pulses can be delivered with great temporal precision. And so we can play in activity patterns as synchronous or asynchronous as we like, at millisecond scale timing, which it turns out to be extremely important. And Viviana's work is a great example because for deep brain stimulation to work in treatment of, for example, Parkinsonism, the frequency does matter; you can't just play in any rate of activity. And there's a specific transition point, if you're in the 20 to 30 pulses per second, you actually can be harmful. And as you get past 70 pulses per second, you transition to the therapeutic. And up around 100 Hz, you've got these amazing results that some patients can experience where you have profound relief of slow movement and tremor. And Viviana was able to use optogenetics to help break down how aspects of Deep Brain Stimulation (DBS) work. And it was critical - the frequency turned out to be enormously important. This timing aspect of optogenetics, which tends to be unique, uniquely enabled by the microbial opsins is something that's showing up again and again. For example, in dissociation, which I mentioned earlier, we had a dissociation paper where the frequency of activity in a particular brain circuit turned out to be crucial. So as you say, that's layered on top of the cell type specificity that we get. And together, we can start to speak the language of brain circuitry. And that of course, matters for the basic science as well as for the therapeutic applications like DBS.
Well, let's get into the therapeutic applications just a little bit more. You've just written a book, which illustrates how your work helps to understand people with mental illness. Maybe you could say a little bit about how your lab work affects your approach to patients, and also how your approach to patients affects your lab work.
Yeah, so the book that you mentioned, it's called projections. It's written for anybody who's interested in the brain. Each chapter starts with a human story, a patient taken from my own personal experience in emergency psychiatry…covering much of the breadth of psychiatry from depression to mania to eating disorders, autism, dementia, bereavement and grief, borderline personality, and so on. Telling that story, I'm able to branch off into thinking more broadly about why this disorder may be present in the human population, and then get into optogenetics and the circuit based understanding of the symptoms - to what extent this has been eliminated by the transformative revolutions that have happened in neuroscience. So it's an unusual book - I took a bit of a risk with it, because it's definitely not a scientific tome. And I think, probably not what a lot of people expected. But what's been exciting has been its positivity, both in the literary critic circles and in the academic community. So a little breath of relief there, despite the risk taken. And it was so important for me to be able to tell these stories. And this is, I think, really helpful for people to know this is a window into the altered state of humanity, and these are shared states across the human family and probably have been with us for a very long time. That connection across the human family and an understanding that now, we can come to a physical material understanding of these states that have been so hard and continue to be so challenging and caused so much suffering - this connection has been really important to me. It's not just a list of checkboxes on some diagnostic criteria list. And that's been so important for the science and I feel so fortunate to be able to forge that link.
So I just wanted to double click to get your perspective about why we even have emotions. Sometimes they really help us out; sometimes they really do not.
Yeah, well, first of all, one way of looking at these internal states we have - the positivity and negativity - this is a way of creating a shared currency within the brain. We have so many competing drives, you know, we've got hunger, thirst, social interaction, parenting, various motivations for other primal drives, you can go on and on. How do you decide at any one moment what's important - these come into conflict all the time? Well, at some point, there has to be a currency exchange, every drive has to be converted into some comparable unit. And so a single discrete unitary action can be taken at any one time. So negativity and positivity of our internal state may be a way of getting to the right decision that evolution and learning has created for an organism. And that is a reason why we don't just act if you go back to this anxiety example, why is it necessary to also feel bad if you're already avoiding the aversive environment? Well, sure, you could just have a reflexive avoidance of the aversive environment. But then when you get into more complex contexts, where you've got a whole raft of competing survival drives at once, it's not so simple. And having that negativity and positivity with the internal state is maybe the way to handle that. But it's some of the most uplifting and wonderful parts of human experience, of course, come with the flip side as well, which is the suffering that we all know.
Let me switch gears a little bit, and start talking a bit more about where psychiatry is going today. And, you know, I think we've seen really interesting developments in the last few decades. One is the rise of deep brain stimulation. Maybe you could take us through these developments.
Well, deep brain stimulation. This, of course, is something that works in Parkinsonism, as we've talked about, it works in a host of neurological disorders. We now have ways we know that it can help people with OCD. We’re starting to get a handle on depression. And it doesn't have to be invasive; there are non-invasive ways of doing brain stimulation like with transcranial magnetic stimulation. But we're not yet armed with the causal knowledge that we need to really use this methodology in a generalizable way for psychiatry, and that's where optogenetics and understanding the circuits, understanding the projections that I talked so much about in the book projections, but also in our over optogenetic work - we're piecing together as a community, the causal material underpinnings of all these internal states. And that's going to be and it is the foundation for all these different stimulation techniques. And the goal is, you know, to move toward a symptom treatment approach where a patient comes with a severe symptom, and we can guide our stimulation armed with causal knowledge and armed with knowledge of the patient's individual brain structure. So that I'm very excited about - there's hope - we're not there yet, in psychiatry and neurology - but the path is clear, because we have this elemental causal approach.
Now, maybe you could say a little bit about the increasing medical uses of psychedelics.
The psychedelic question is extremely interesting. First, just a little bit of terminology, there are a number of mind-altering agents that you'll hear discussed. For example, ketamine, you'll see discussed in the context of depression. It's a very interesting agent that also causes dissociation. And we've studied that. That's not a psychedelic, but it's certainly a next generation mind-altering substance. There's also MDMA, which is more of an empathogen, not technically a psychedelic either, but as being productively explored as a strategy to put the brain into a new state and allow it to be more receptive to and to modify itself more with certain types of therapy. And then you've got the true psychedelics hallucinogens that directly cause altered perceptions. And this would include things like LSD. Now, first of all, I'm as interested as anybody in this sort of work. In my own lab, we work with all these - we've published on ketamine, we've published on MDMA, we’ve published on LSD - these cause such remarkable, profoundly interesting effects on sensation, perception and action that any rigorous brain scientist should be - and they are - intrigued by what these are, what they're doing. And then, of course, the natural question is, can we use these for psychiatry? There are risks, these are agents that are not without side effects, not without addiction risk, not without the possibility of causing long-lasting, you know, potentially irreversible negative effects on the brain. And so we have to approach this very cautiously. But here again, this, this theme we've had in our this wonderful conversation of breaking things down to their parts is so valuable, because by studying what these agents do, and getting to a deep understanding of the circuit, effects of these agents will help us hone in on which are the therapeutic ones and which are the harmful ones, and will help us identify ways to selectively recruit the helpful effects. So I'm optimistic; you know, we have to be cautious, but the path is very, very intriguing.
And then also some breakthroughs in basic biology about the understanding of synaptic pruning.
Yeah, there's a number of recent studies over the past few years, suggesting that a number of disorders relevant to psychiatry, including dementia and schizophrenia, that there are these extremely interesting brain immune interactions or actions of traditionally immune molecules that are present in the brain that cause structural changes that cause physical reshaping of circuits. And this is not just observational; it's buttressed by genetics. So you can see that, for example, in schizophrenia, that elements of the complement cascade, which is part of traditionally blood-related set of molecules. It turns out to be pretty important in how the fine structure - the spines and dendrites that were synapses formed connections formed from one cell to another in the brain - these can be reshaped for good or ill by these traditionally not-brain-associated molecules. So this is again early days, but extremely intriguing. And it shows us there's always surprises that nature has in store for us. If you look more broadly, modern technology has let us see both the forest and the trees at the same time. And so this is the advent of truly brain wide, but also high resolution, observation. And we've built for example, very wide field optics, we call it in lab, that let us see across the whole brain at once and see patterns and rhythms in a cell type specific way. But without sacrificing the globality, we can see individual kinds of cells and even individual cells across the full sweep of the brain. And that's led us to identify principles that we couldn't have discovered otherwise. And that's come through advances in optics in building new kinds of microscopes. So if you look at all these advances together, there's not much that's more exciting than being a neuroscientist today.
That's absolutely fascinating. And I definitely share your excitement. I want to add on another aspect that I see emerging: behind all this optogenetic stuff is computer control, right, really fine programs that are able to do quite fine time scale manipulations of the brain, and in some cases, in closed loop. But outside the lab in the world, we're constantly exhausting information through our mobile phones through our computers. And there's an enormous amount of information about our internal state that's available, through how we behave, how often we charge our phones - something as simple and prosaic as that. And so you might call this digital psychiatry or wearable technology, and maybe you'd even add them the buzz terms of “AI” to help analyze this information. And we're starting to see that right. So AI analysis of speech can detect depression more accurately than some practitioners. So do you see yourself using digital psychiatry writ large in the future for your work?
Yeah, I'm intrigued by this. It has a plausibility to it. And I talked about this in the book quite a bit as the experience of a psychiatrist forging a connection with a patient and realizing that it's not all completely dependent on the full human interaction of face to face, in person. And I talk about my experiences as an on-call doc at the VA hospital where patients would call in, and all I would have was this audio stream. And I found the importance and the intensity of these moments could hardly be overstated, because these were veterans calling in because they were close to suicide. Maybe they're suffering from PTSD, as they often were. They're experiencing flashbacks; they're experiencing severe depression or panic. Many of them are close to suicide, unfortunately, some do come to suicide. And with enough experience, just the rhythm of the sound, just the rhythm of the voice, I could start to realize this is the level of depression in this person, this is the level of concern I have to have. This turned out to be really useful in the Coronavirus epidemic time because now we're seeing patients with reduced data streams all the time - patients can't come in. We're seeing them by audio, we're just seeing them by video without the full interpersonal interaction. And I'm reactivating those old skills that I learned from my VA hospital time with those phone calls. And I'm realizing, yeah, it's still there, I can still do it. So from where I sit, it's certainly plausible that the wearables and the AI data streams contain rich information, but only on one part. They can reflect the whole and they can carry meaningful information that relates to how the human being is doing. Of course, as we all know, a big risk of AI is getting it wrong - coming in with a poorly trained system or a system that wasn't trained in the right way, or that countered an adversarial example, and so on. And so we have to, just as with psychedelics, we've got to be careful. But I'm optimistic. I think this will, just as the computer program-based calling of EKGs is helpful for cardiologists, this will be an important part of psychiatry in the future.
Karl, thanks for the conversation for your amazing perspectives and sharing your history and your science.
Thank you so much, Karl, really a pleasure.
Wonderful, thank you. It's been a lot of fun.
We usually take time at the end of each episode, in the spirit of regular in person meetups in Boston many years ago, to discuss a big problem: the nail and possible solutions; the hammers, inspired by what we just heard. Alex, what do you have this week, a hammer or a nail?
I think it's a hammer. I think also, as we get further along in this series, sometimes the hammer and the nail distinction becomes a little bit blurry. But what he wanted to talk about today, and we did touch on it a little bit in our conversation with Karl, is this idea of digital psychiatry and data exhaust. So when we're using our phones, when we're using our computers, when we're surfing the web, when we're plugging in our phones, a lot of that information is stored. It's logged and it's used right to improve the experience on the phone, or it's better battery usage in the case of how often you charge it, and better search results in the case of previous search activity, things like that. I mean, we live so much of our lives online. And a lot of that information is being used in order to improve those experiences and to make our devices in our lives flow more smoothly. Now that information is used in a consumer context and kind of a company context and like a social life context. But there's this idea that's emerging. And I'm not an expert. I'm not a psychiatrist. I'm just a caveman, a former neuroscientist who also likes computers. But what's emerging is this idea that that data might actually be useful for understanding and improving our mental health.
Oh, yeah, it's super interesting. So tell me a little bit more. I mean, who are some of the people doing this research? And what are some of the initial insights?
You know, I think the first result that I saw that sparked my interest in this was paper by Barabasi, and this is more than a few years ago. And what this paper showed is that, by just looking at what cell phone tower your cell phone was connecting to, and looking at the sequences of those different towers that you connect to back in the times when we actually traveled to and from work - it was clear that people actually lived relatively stereotyped lives. It's not a surprise, right? I mean, you get up in the morning, you go to your favorite coffee shop, you go to work, maybe you go you know, for drinks with friends, you come home, right? And then Saturday and Sunday are gonna be different, right? So the stereotypy of the rhythm of our everyday lives is apparent in the log of what cell phone towers you connect to,
I think this was the paper that was entitled something like “Lévy flight.” And one of the interesting insights was that if our distribution of motion was a Brownian motion, then the distances we travel end up being roughly normal, whereas what you see is the distances that I travel, actually follow more of a power law, where most of the time I take little tiny distance traveled. But then every so often, I fly to another country. And then that's like, extreme far off travel. And so it was this different stochastic process that describes our emotions, if I recall correctly.
Yeah, Lévy flight and Brownian motion and these kinds of statistical descriptions of motion - what's wild is it applies to us. At least, I believe that I'm a complex person with complex desires. But I live a pretty habitual life. And it's now reduced in complexity due to Covid. The insight is that information like this can tell us when we're behaving normally or when something is going wrong. So if you enter into a depressive episode, where you spend your time, your ability to get to work on time, what you do on the weekend - will change. And it might be revealed in data exhaust - the logs of the information, and the traces that we leave with our digital devices. So the second piece of work that was really inspiring, revealing to me, was by Jukka-Pekka Onnela. I think actually, you told me about this, Anthony. The insight is, if you just look at how often people charge their phone, it will tell you whether or not you are in a depressive event. The way in which that insight was arrived to, is basically recording as much as possible from a person's life online and on their phone. People opted in, of course, and there was, I believe, software that recorded the number of keys you'd hit on the keyboard per minute, and the kinds of social networks that you've formed through texting and all kinds of other things. But one of the most predictive features is: do you plug in your phone at night regularly? And people who don't do that, who have disrupted habits around their phone, which indicates that it might be an almost ritualistic type device to us these days, is highly indicative of depression. And, that to me, was really surprising. And there are other companies that are looking at this - other researchers who are looking at this - a friend of mine, Justin Baker, is a big proponent of this perspective, and he's now at Mindstrong. Mindstrong - it's a company that has a perspective that what you use your phone for, how you use it - will be diagnostic of different mental states. Things like - how dexterous are you at using your phone's keyboard - might actually contain signal for whether or not you need to be looked at by a psychiatrist, which I think is a really interesting concept.
Alex, you should say a little bit about the work that you did as a PhD student with videotaping mice and doing phenotyping.
Yeah, I think a lot of my perspective was developed in the process of thinking about this stuff. So I used to be a neuroscientist who worked with mouse behavior, then I developed a pretty severe mouse allergy. And so I don't do that anymore and got more into computers as a result. Typically what you do for mouse behavior is you either physically restrain them, and you need to do that in order to use microscopes to view into the brain or to do certain kinds of brain recordings, or you train them so that anytime a light flashes, or a sound beeps, they do something very stereotyped because that's how they get their juice or get their food. And then you can study their brain and behavior. As a result, what I worked on - this was with Bob Datta at the Harvard Medical School - was JP Onnela’s perspective, which is, what if we just record everything that the mouse's body does, and what if we record it with a 3D camera, so we actually can see all of the little different changes at the millimeter level, as it twitches in bounces and sniffs around a box or, or a bucket? And turns out, that's pretty hard to do. Because you’ve got to take this 3D image of a mouse as it's moving, and you’ve got to segment the mouse, and then you’ve got to align the mouse and then you’ve got to cluster its behavior. But we managed to build that whole system up together, and then get a time series of the shape of the mouse's body. And when you look at that data, it kind of looks like spoken language. And the problem kind of looks like parsing natural speech, and parsing sentences into words and words into syllables. This work was done very, very closely with Matthew Johnson, who's now at Google and with Ryan Adams, who's at Princeton. But basically, the idea was that body language is language, in the same way that the spoken word is behavior - its behavior of muscle contractions, and the behavior of our vocal cords, that very conveniently, is turned into a one dimensional signal, which is pressure waves over time. Behavior is exactly that - it's just more complicated because it's not something that you can easily capture and turn into a one dimensional signal.
Give me a sense of like, the size of this vocabulary, and what does its grammar look like?
How many letters are in the alphabet of a mouse's body language? It turns out using the perspective and methods that we have, the answer is about 60. And it always kind of came out to about 60. But when the mouse is behaving in a laboratory setting, there's about 60 unique things that it can do with its body. And you might ask, how long did those letters last? Like? Is this a second long behavior or a minute long behavior? On average, they're about 300 milliseconds. So they're very short. It's about the duration of a mouse step. It's an integer multiple of their average heart rate. Oh, wow. And there's really interesting 300 millisecond rhythms that occur in the brain. And that's about the time it takes for a signal to make it from the basal ganglia to the motor cortex back to the basal ganglia. So there's these three important loops, neural circuits in the brain that seem to operate at this fundamental frequency?
Well, the basic question is, what would it take to do this in humans? I mean, for example, could you run this on all of YouTube videos, and try and parse out the atoms of human behavior from that?
Yeah, so we tried this a little bit at the tail end of my PhD for human faces. And what we did is we took pictures of people, and there was an alignable event - that's a really important thing for us; there's a thing that was happening. There are databases that include people who are rotating their shoulder, and they have a shoulder injury. And so above a certain point starts to be painful. And this is just in the normal course of assessing range of motion. But you know, at a certain point, it starts to hurt. And what we did is we analyzed the key points of these people's faces, and put that as an input to our behavioral analysis programs, and discovered the number of “facial syllables” that are available from the data that we have of these people's faces. And it turns out, if you use that as a feature relation, or an embedding of people's faces, you can use it to predict the pain score.
And you know, there's different ways of reading pain or grimacing. And this approach is very predictive of that. The key difficulty is in turning an image or a video into a set of numbers that really does represent the current pose, and changing pose of an animal. It's hard to do that for us. Because we can walk on all fours, we can hop on one leg - pretty challenging problem - but there's lots of people who are actively trying to work on feature-rising behavior. And that's the first step to understanding its components.
This is fantastic. Alex, I could talk to you all day about this.
Let's do that sometime.
Thanks so much for this discussion today.
Absolutely, it's always a pleasure.
Huge thanks to Professor Karl Deisseroth for joining us today. Next episode, we'll be meeting Geoff Hinton, the godfather of deep learning, to talk about how humans and computers learn, and Cynthia Kenyon to talk about the molecular and cellular basis of aging. If you have any questions for us, or our guests email firstname.lastname@example.org or tweet at @GVteam. We'd love to hear from you.
This is a GV podcast and a Blanchard House production. Our science producer was Hilary Guite and executive producer Rosie Pye, with music by Dalo.
I'm Anthony Philippakis.
I'm Alex Wiltschko.
And this is Theory & Practice.