76 minutes
SDS 547: How Genes Influence Behavior — with Prof. Jonathan Flint
Subscribe on Website, Apple Podcasts, Spotify, Stitcher Radio or TuneIn
Dr. Jonathan Flint, Professor of Psychiatry and Biobehavioral Sciences at the University of California Los Angeles joins us to discuss how he uses data science and machine learning to explore the link between genetics and depression. Learn about the role that ML plays in modern genetics research and discover how Johnathan is partnering with organizations around the world in UCLA’s The Depression Grand Challenge.
Follow Jonathan:
Thanks to our Sponsors
About Jonathan Flint
Jonathan Flint is a British behavior geneticist and Professor in Residence in the Department of Psychiatry and Biobehavioral Sciences at the David Geffen School of Medicine at UCLA. He is also a senior scientist in the Center for Neurobehavioral Genetics at UCLA's Semel Institute for Neuroscience and Human Behavior. He fundamentally advanced understanding of the genetic basis of behavior, thereby determining the direction of research in psychiatric genetics. He was elected as Fellow of the Royal Society in 2019.
Overview
In this episode, Jon reunites with his former Oxford professor Dr. Jonathan Flint to discuss his role in UCLA's The Depression Grand Challenge–a campus-wide initiative aimed at cutting the burden of depression by half by 2050.
To begin speaking about depression, Jonathan first identifies the complex link between genetics and behavior. Every human behavior is associated with variants, and by identifying these variants, one can have a better handle over them. These variants can either be genetic, environmental, or gene-by-environment factors. In particular, he highlights that the single disease that we call depression today is likely to be many different diseases, each with its own distinct genetic basis. "Depression is not one disease," says Johnathan, which leads him to identify a vital machine learning problem: "can we make predictions from all the genetic signals we are collecting to recognize subgroups?"
But what about scale, Jon asks? Suppose you want to make a robust detection of one small signal. In that case, you need to analyze an enormous sample size of people. The Depression Grand Challenge operates on a nearly unimaginable scale. Each of the genes affecting depression contributes to such a slight variance in whether someone is depressed or not. And the study has potentially tens of thousands of these genes interacting across millions of positions across the genome. What's more, specific variants may hold more weight across cultures and populations around the world – just another factor that needs to be taken into account.
Johnathan's team also works with audio engineers, who can extract over 700 features from the interviews they conduct. This very high-dimensional data set, coupled with the genetic and environmental factors, generates an enormously complex data set. Ultimately, Johnathan expects to use this detection technology at hospitals during the patient interview process and hopes it will help detect depression and potentially save lives.
For those wanting to help, Jonathan insists that the first step begins with listening and reaching out to those experiencing low moods and asking if they're contemplating suicide. And if you're interested in working directly with Johnathan, simply sending him an email is always a great start.
The pair then tackled a few audience questions from LinkedIn.
After revealing that southeast Asian countries have the lowest level of depression, Jonathan also shared the cultural differences that significantly impact the way the data is collected, and even which questions are asked. "There's a whole other science around how you acquire information from people," Johnathan says.
And as far as what excites Jonathan about the future, he reveals their latest partnership with Apple and their work with Apple Watch data. UCLA has now partnered with Apple for testing sensor technology on the Apple Watch.
In this episode you will learn:
- Johnathan's background [2:53]
- How we know that genetics plays a role in complex human behaviors including psychiatric disorders like anxiety, depression, and schizophrenia [8:00]
- The role that data science and ML play in modern genetics research [15:08]
- About Jonathan book "How Genes Influence Behavior" [19:45]
- The day-to-day life of a world-class medical sciences researcher [32:24]
- The open-source software libraries that Jonathan uses for data modeling [40:33]
- A single question you can ask to prevent a severely depressed person from committing suicide [52:00] LinkedIn Q&A [54:41]
- The future of psychiatric treatments [1:05:35]
Items mentioned in this podcast:
- How Genes Influence Behavior by Jonathan Flint, Ralph J. Greenspan, Kenneth S. Kendler
- UCI Master of Data Science Program
- Monte Carlo
- Julia
- Python
- R
- Contrastive Principal Component Analysis
- Hidden Markov Models
- UCLA Depression Grand Challenge
- Quantum Computing Since Democritus by Scott Aaronson
Podcast Transcript
Jon Krohn: 00:00:00
This is episode number 547 with Dr. Jonathan Flint, Professor of Psychiatry and Biobehavioral Sciences at the University of California, Los Angeles.
Jon Krohn: 00:00:13 Welcome to the SuperDataScience Podcast, the most listened to podcast in the data science industry. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. I'm your host, Jon Krohn. Thanks for joining me today and now let's make the complex simple.
Jon Krohn: 00:00:44
Welcome back to the SuperDataScience Podcast, you are in for a special treat today with Professor Jonathan Flint, one of the world's leading intellectuals on the genetics of behavior. Jonathan is professor in residents at UCLA specializing in neuroscience and genetics. He is leading a gigantic half-billion-dollar project to sequence the genomes of hundreds of thousands of people around the world in order to better understand the genetics of depression. Originally trained as a psychiatrist, Jonathan established himself as a pioneer in the genetics of behavior during a 30-year stint as a medical sciences researcher at the University of Oxford. He has authored over 500 peer-review journal articles and his papers have been cited an absurd 50,000 times.
Jon Krohn: 00:01:31
He also wrote a university-level textbook called "How Genes Influence Behavior", which is now in its second edition. In this episode, Jonathan details how he know that genetics plays a role in complex human behaviors, including psychiatric disorders like anxiety, depression and schizophrenia, how data science and machine learning play a prominent role in modern genetics research and how that role will only increase in years to come, the opensource software libraries that he uses for data modeling, what life is like day to day for a world-class medical sciences researcher, a single question you can ask to safely prevent a severely depressed person from committing suicide and how the future of psychiatric treatments is likely to be shaped by massive scale genetic sequencing and everyday consumer technologies. Today's episode mentions a few technical data science details here and there, but the episode will largely be of interest to anyone who's keen to understand how your genes influence your behavior, whether you happen to have a data science background or not. All right, you ready for a deeply fascinating episode? Let's go.
Jon Krohn: 00:02:43
Jonathan, welcome to the SuperDataScience Podcast. I can't believe you're here, this is such an exciting moment for me.
Jonathan Flint: 00:02:50
Jon, I'm very pleased to be your guest today.
Jon Krohn: 00:02:53
So, Jonathan was my PhD supervisor at Oxford, so 10 years ago, I finished my PhD and so, I guess I've seen you in person maybe once when I defended my dissertation maybe about eight years ago, but I can't remember for sure if I actually saw you then. And I had plans to see you in LA because the company that acquired my company was we have an LA office, but then the pandemic hit. And we even had plans, I was going to see you in the autumn, we had specific plans and my flight was canceled, everything because Covid kicked back up in LA. But anyway, so you're not in Oxford anymore, you're in LA, how's it going out there on the West Coast?
Jonathan Flint: 00:03:36
It's great here, Jon, wonderful lifestyle, great colleagues, great place to live.
Jon Krohn: 00:03:41
So, you don't miss the sun going down at 3:00 PM and it raining all the time?
Jonathan Flint: 00:03:46
I don't miss the overcast days, the constant temperature at about four degrees centigrade. I miss none of that.
Jon Krohn: 00:03:55
You also in Oxford, you would cycle everywhere all the time. Are you still cycling everywhere in LA? You must drive, right? Yeah.
Jonathan Flint: 00:04:01
[crosstalk 00:04:01] No, no, I-
Jon Krohn: 00:04:01
Oh, you are? You're cycling there?
Jonathan Flint: 00:04:02
Yeah, I cycled in today, I cycle in every day.
Jon Krohn: 00:04:05
Wow. All right, is that a long commute for you?
Jonathan Flint: 00:04:07
It's about six miles.
Jon Krohn: 00:04:10
Oh, that's not bad at all. It's nice.
Jonathan Flint: 00:04:12
No, it's not bad at all. No, it's a perfect cycling city. It's always sunny here, the streets are flat and the cars here have never seen a cyclist so they behave very sensibly and they give me right of way so it's good.
Jon Krohn: 00:04:25
Wow. All right. Wonderful. So, you're at UCLA and you're tackling something called the Depression Grand Challenge, which sounds grand indeed. It is a project that's going to take several decades to carry out and it plans to raise over half a billion dollars in funding. So, how did you get involved in this grand challenge? Why are you doing this?
Jonathan Flint: 00:04:50
Okay, so let me go back a little and just explain why I'd been involved in this at all in the first place. And I'll tell you a little bit about myself and my background, I think that gives you the necessary context with this. So, when I was a teenager, my mother was working for an organization called the Samaritans, which was the first, certainly in the UK, maybe in the world to organize suicide helplines. And during one grim Christmas, when everyone has eaten and drunk too much and [inaudible 00:05:25] to talk anymore, my mother said why didn't I accompany her to go into the center of London and man one of the suicide lines.
Jonathan Flint: 00:05:31
So I was like, "No way, I can't do this. No, you're just going to be dangerous. I'm going to be really unable ..." and she said, "No, no, no, no, I'll be with [inaudible 00:05:40]." So, she convinced me and I came and I sat in a crypt of a church in the center of London and there's a bank of telephones in front of us. And one of them rings and I pick it up and my mother looks at me, "You can do it." "Oh okay, I'll take the call." And at the end of the phone, there's a woman who tells me that she wants to kill herself, that she has the pills in front of her and she talks and she talks. And at the end, she says, "Thank you for listening, I'm not going to kill myself, I'm going to throw away the pills."
Jon Krohn: 00:06:12
Wow.
Jonathan Flint: 00:06:14
And for the first time I realized that I could do something, that even listening is something. And in fact, that's something we can all do. It's not a huge problem that we can't confront. No, we can all get involved in this. So as you know, I'm a psychiatrist and being a psychiatrist, there's lots more stuff I can do. I can give pills and I can give psychotherapy to my patients. But fundamentally, that approach hasn't changed. And I think we can all get involved in this. It's a problem that's extremely prevalent. Almost everybody that haven't suffered themselves will know somebody who's involved in this condition. And it's so broad, touching so many different parts of our lives that you really need a joined-up approach to tackle this. Now, every department of psychiatry in the country will have people working on the causes of depression and trying to improve things but that initiative is not enough, you need something like the National Institute for Depression. And Australia has got closest to doing that but in the US, the one place that really caught my imagination was UCLA. And they tackled this, as you've said, as a grand challenge.
Jonathan Flint: 00:07:31
In other words, something where we could make a transformative change in a major problem in society. And there was a series of meetings around how to do this and I attended those meetings and they ended up offering me a job and how could I resist? This is the problem that I've been tackling all my life and they were doing it on the right scale. So, I came. And we've been running now for about five years and it's an uphill struggle, no doubt about that, but we are making progress.
Jon Krohn: 00:08:00
And so, the idea here is that there is some genetic basis to depression and that by understanding this genetic basis, we might be able to treat better.
Jonathan Flint: 00:08:12
So, okay. So, my training is as a psychiatrist so we use genetics as a tool on a general scale but we can talk about this a bit more. Every human behavior, in fact, almost all parts of physiology, everything varies. So, we all know height, weight vary, so does mood. And because of the way biology works, there will be genetic variants, which are associated with that variation. So, if you can find those genetic variants, it gives you a handle on the condition because you then got to handle through the genetics of the biology. Now that's not to say, and many people misunderstand this, that the environmental effects are not important, they are hugely important, particularly in something like depression. And there are many people who become depressed because of something bad happening to them. That's an environmental stressor. But that's not to say that the genetic effects are also not important, even in those circumstances, because we know that you need to have a certain degree of genetic predisposition before an environmental stressor will really upset you. And we all know people who've survived catastrophes unscathed. Why does that happen? How do we explain this huge variation? It's not just down to one genetic predisposition, there are other things going on that contribute to this.
Jon Krohn: 00:09:29
Right. So, for any trait or any even physiological trait, behavioral trait that an organism has, that a living thing has, including people, some of the influence on that is genetic, some of it is environmental and that means, basically anything that isn't genetic, right?
Jonathan Flint: 00:09:46
Yes, that's essentially what it means. Yeah.
Jon Krohn: 00:09:49
So, every single stimulus that you encounter over your lifetime, this is some kind of environmental influence. So with depression, for example, it could be things like socioeconomic status is an environmental factor and that's the kind of thing that you can study. Okay, what was the socioeconomic status and put the person in a bucket, but then in practice as an individual, it's literally every single thing that's ever happened to you from the instant that the egg was fertilized by a sperm, in utero, as a child, as an adult, all of these experiences are your environment.
Jonathan Flint: 00:10:30
Correct.
Jon Krohn: 00:10:32
So you mentioned, so we have genetic influences on any trait, including depression, susceptibility to depression, severity of depression, environmental factors influence it. And then, you mentioned this gene by environment interaction as well. And so, this idea that two people with different genetics exposed to the same environmental conditions, which of course in practice, you can't ever have exactly the same environmental conditions as somebody else. But if you could, then people with different genetics would react differently to that environment. And one could become depressed, the other not, just as I suppose, one could become taller than the other in different kinds of environmental conditions. So, we have a genetic component to any trait, including depression, an environmental component and this gene by environment component. And if I remember correctly from my PhD, which is now a decade passed, there are studies from twins raised apart and I think Kenneth Kendler is a big name in these kinds of studies. So, you compare how do identical twins-
Jonathan Flint: 00:11:44
You don't need to have twins reared apart, I mean, that's a very rare phenomenon. I mean, Ken's work has always been on-
Jon Krohn: 00:11:50
But I thought that provided ... Oh, okay, I thought that provided some of the richest data when you had-
Jonathan Flint: 00:11:55
Well, the twins raised apart, those studies have been much advertised because they are so unusual and Tom Bouchard has run this for many years to find these people. But the trouble with it, there are so few of them that you don't really get a good estimate of some of the parameters that you want to look at. I mean, the reason that they got the people's attention was because it was hard to argue that there was something special about the environment the twins were raised in because they were obviously, they were separated. And then people argued, well maybe early on they were together and that was what's important and how important were the early ... So, I think one can put that aside and just say, "Twin studies by themselves are pretty informative."
Jon Krohn: 00:12:38
Nice. Okay. And then so, it's then the difference in behavior, in some trait like depression, between identical twins versus fraternal twins that the differences matter and it doesn't matter whether they were raised apart or not so much.
Jonathan Flint: 00:12:53
No, not so much. This turns out to be quite a rather specialist sample. But yes, so there are twins who are genetically identical so therefore, any differences you see between them have to be environmental. There are few little wrinkles on that but as a general rule, that's a good starting point. And then, there are twins who are like brothers and sisters, so they share half their genetic material. And by comparing the two sets of twins, you can come to some estimate of the extent to which genetic variation contributes to the heritability of a trait.
Jon Krohn: 00:13:24
Right. Okay. And then so, if I remember correctly as kind of a general rule of thumb, for a lot of complex traits, including depression, about a third of the variation in depression is genetic, about a third of it is environmental and then the remaining third are these gene by environment interactions.
Jonathan Flint: 00:13:42
We don't really have a figure on the contribution of the gene by environment but roughly it's about 35% is heritable and the rest we say is environmental, which essentially means we don't know what's going on.
Jon Krohn: 00:13:56
And there's too many possible influential factors and we can't control for all of them, about 35% genetic, okay.
Jon Krohn: 00:14:06
Many thanks to the Master of Data Science Program at the University of California, Irvine for sponsoring today's episode, the UC Irvine Master of Data Science Program blends statistics and computer science principles with partners from industry to empower students to innovate in the field of data science. Located on the tech coast of Southern California, students of the program will enjoy a powerhouse ecosystem where over a third of Fortune 500 companies are located. Take a giant leap in your data science career through the UC Irvine Master of Data Science Program. To learn more, head to superdatascience.com/uci. That's superdatascience.com/uci. Check it out.
Jon Krohn: 00:14:52
So, how does understanding that say, a particular gene is likely to be causal to some extent of depression. And I know that any individual gene contributes a small amount. But let's say, let's oversimplify for a second and say that you could find a gene that was hugely responsible for whether somebody had depression or not. What could we then do with that information? Why is that helpful?
Jonathan Flint: 00:15:21
Okay. So, we need a bit of context for that and I think the starting point really has to be, to think a little bit about what depression is and what it isn't and making a diagnosis is not that straightforward. Like if I'm talking with you now and I ask you, "Jon, have you felt in the last couple of weeks a bit down?" You'd probably say, "Yes, there's been a few times," and so sometimes when you haven't slept too well and you probably say yes and I could elicit without much difficulty many of the symptoms that will be diagnostic of depression. So, there's a bit of an art, clinical art in really establishing whether someone has got depression, so that's the starting point. And when I ask those questions of you, I'm drawing upon decades of experience from clinicians who try to work out what it was that characterized depression, not because it comes from any understanding of the underlying neurobiology, we're just asking clinical questions. So at that level, you can think of it as a syndrome, it's a set of features. And what's unusual about depression, probably almost unique, is that you could have completely opposite traits and still qualify. So, if you are sleeping very poorly and getting up early, that will feature, but you may also be sleeping longer. You may be losing your appetite and eating less, but you may also be eating more than usual.
Jonathan Flint: 00:16:47
And weirdly both of those, I mean, both of those can increase your score for being diagnosed as depressed. So, given that, it's not, I think too much of a stretch to realize that depression is almost certainly not one disease. And that when we deal with it, it's a bit like we were dealing with cancer but we didn't know there's a blood cancer, that there's a lung cancer, all we know is that some abnormality of the cells and they keep on dividing. So, it's a bit like that with depression that we have this phenotype that the clinicians have difficulty eliciting and we know that there's some genetic component. But really beyond that, we are pretty much at sea. So, one of the first questions we want to answer is, is it one disease? How many diseases and what are the characteristics of those?
Jonathan Flint: 00:17:38
And here, the genetics becomes very useful because if it's not one disease, then you would expect there to be different constellations of genetic factors operating. Now this becomes, guess what, one level of machine learning problem, can we make predictions from all the genetic signals we're collecting to recognize subgroups and so on? So, the way we think about this is that genetic, you will know this but many people won't, that in common with other complex traits, we regard it as polygenic, hat's to say, there isn't a single gene, there's not two, there's not three, not even 10, not 100, no, there are thousands, probably tens of thousands of genetic predispositions spread across the genome. And each of those makes a tiny, tiny contribution, each increasing your risk by a minuscule amount.
Jonathan Flint: 00:18:26
Like if you put this as an AG ratio, 1.01 would not be unreasonable. If we find things that are 1.2, 1.3, we're really, really happy, but it's the majority of them are going to be really tiny, tiny effects. And we'll come to what those understanding those individual effects might tell us and that was really your question about genes. But I think before we get to that point, we can use the genetic data in a way to answer this more fundamental question which is, are we dealing with a single disease or multiple diseases? Because if you didn't distinguish the group and then you found what you think is the causative gene, let's say, and you work on what it's doing, it may be that that's something completely unrelated to many of the conditions that we're actually interested in. It might be some morbid feature, it might be some consequence, it's very hard just on that single starting point to understand what the gene would be doing. So, that's a bit of a long answer, but the question you ask, Jon, many people ask that, but I think to get to the root of this, you got to step back a little bit and understand what the condition is.
Jon Krohn: 00:19:34
Yeah, that was a really great answer. All right. Thank you, Jonathan for that overview of how genetics contributes to behavior, especially to depression. Now, if people want to learn a lot about this particular topic, there's a great book recommendation that I have for them and that's "How Genes Influence Behavior" and huh, you just happened to be the author of that book.
Jonathan Flint: 00:19:57
Wow, what a surprise.
Jon Krohn: 00:20:00
So, the second edition came out in 2020 so quite up to date. Jonathan, what do you cover in that book and who's the intended audience?
Jonathan Flint: 00:20:10
Well, this book started as a collaboration, set of conversations between myself, my colleague, Ken Kendler and Ralph Greenspan. And Ralph is a Drosophila geneticist, he works on flies. Ken's only ever worked on human genetics and I crossed the boundary between mouse genetics and humans. So, we came at this from different angles and our main intention was to try and both educate and we hoped also entertain. We'd originally written this for fun as we thought it would be important to get our message out and we didn't have a particular audience in mind. But then, when we went to a publisher and tried to get someone to publish it, they asked that question and we said, "We don't know." So, we ended up working with Oxford University Press, who has published it as a textbook for undergraduates and first year postgraduates in psychology and psychiatry.
Jon Krohn: 00:21:12
Nice and yeah, so roughly what does it cover in the beginning? I suppose it-
Jonathan Flint: 00:21:20
We wanted to cover both the phenomenology of psychiatric disease together with what is known about its predisposition and also how from model organisms, we have information about how genetic effects can binge up on behavior. Because obviously experimentally, there's much more you can do with mice and flies. And there are lessons from those experiments to teach us what is the answer to the question we pose, namely, how do genes influence behavior?
Jon Krohn: 00:21:50
Yep. And so yeah, of course with fruit flies, Drosophila, they breed very quickly. I imagine it's maybe every few days.
Jonathan Flint: 00:21:59
Indeed, yeah. We can get lots of flies quickly, yeah, we have to keep an eye on them.
Jon Krohn: 00:22:02
What are guys up to in there?
Jonathan Flint: 00:22:05
Well, it's more that the other fly geneticists have to go off to check on their virgins because they don't want get them inseminated unnecessarily and so ...
Jon Krohn: 00:22:15
That's the day to day life of the-
Jonathan Flint: 00:22:16
That's the day to day life of the fly geneticist, yeah.
Jon Krohn: 00:22:20
Surely, well I guess, yeah because I would say, well you just keep them separate but that's hard if they're breeding all the time.
Jonathan Flint: 00:22:25
Correct. Yeah, not so easy. Ah, now you begin to realize.
Jon Krohn: 00:22:29
And fruit flies are so tiny, they're just a few millimeters long so you've got to, how do you ... you get in there-
Jonathan Flint: 00:22:33
We have a little sucking tube and we suck them up and put them into a [inaudible 00:22:39].
Jon Krohn: 00:22:37
It's like an eyedropper.
Jonathan Flint: 00:22:40
Yeah. If you want, hold on a second I can just show you, this is getting a little technical, but maybe your audience will be interested in this. This is fly pushing, it's Ralph Greenspan's, but it answers all the questions that you might ever want to know about actually how to handle the flies, how to breed them, how setting up across, it's all in here. It is very, very, very hands-on technology because one of these books has got one of these wire binders in the middle, so opens on your bench while you're doing the experiments.
Jon Krohn: 00:23:13
Oh yeah. And so yeah, so this is probably a surprise to a lot of listeners that fruit flies are a key model animal in the study of behavior. And while there are certainly a lot of differences between a human and a fly, it's surprising in terms of how our biochemistry works, in terms of how we use energy or consume nutrients, there's still a lot of common underlying genetics. So yeah, in terms of just how our cells behave, there's all of these underlying commonalities. And then, even surprisingly for something that we think is quite complex like behavior, there's still commonalities, aren't there Jonathan, between behavioral patterns and a fly and a person?
Jonathan Flint: 00:24:06
So, the key person here was Seymour Benzer who in Caltech decided to take the power of fly genetics and apply it to behavior. Much berated at the time, people said that wasn't possible, it was fine to do it with physiology and metabolism behavior. And he set up experiments to try and identify genes involved in learning and memory, which eventually was successful and to everyone's astonishment, the genes that he identified turned out to be the same genes in mammals, including you and me.
Jon Krohn: 00:24:33
Wow. And then so, as an example of something that we can do with fruit flies that you couldn't do with a human and that would probably even be trickier with a mouse, is there are, well at the time that I was in Oxford this was really cutting edge, was changing the genetics. So, inserting genes into the fruit fly genome so that you could activate particular neural pathways, particular behaviors by shining a laser on the fruit flies.
Jonathan Flint: 00:25:06
Yes, you're right. I mean, this is technology that my colleague [inaudible 00:25:10] developed, it is available in mice so that the whole approach, which is called optogenetics has really transformed neurobiology because you can, as you said, go in and activate a part of the mouse's brain, part of the fly's brain and test how brains function. You can do experiments on what you think neurons might be doing.
Jon Krohn: 00:25:33
Cool. Yeah. And obviously, that doesn't get past university ethics research boards when you want to do it to people.
Jonathan Flint: 00:25:41
We can't do that on people yet, but on the other hand, our abilities to use computers to allow us to activate different parts of the body so you can try and get people with spinal cord injuries to overcome their disabilities has made huge leaps and bounds.
Jon Krohn: 00:26:01
And deep brain stimulation.
Jonathan Flint: 00:26:03
Deep brain stimulation. So, we're not that far off.
Jon Krohn: 00:26:09
Cool. So, a question that I have related to this is, so we have this idea, okay in fruit flies, we can get lots of data, we can do lots of experiments on them because the fruit flies breed more quickly so it naturally allows us to get a lot more genetic variability. With people, we don't breed nearly as often, we're not having a new generation every few days, takes decades typically. But to study people like you do and try to identify the cause, the genetic causes, the genetic factors behind some complex trait like depression, what kind of a scale do we need to be working at to identify these genes?
Jonathan Flint: 00:26:58
So, remember we talked a little bit before about how the genetic effects on behavior and depression as I'm interested in, but this is true for diseases like schizophrenia as well as other medically important traits. The genetic effects are tiny so each gene variant is making a very small contribution. In order to detect that, you need a very large sample size. This is just a standard power problem and if you want a really robust detection of a small signal, you will need to analyze a large number of people. And to put this into further context, we are interrogating entire genomes. So, the genome and three billion bases of it has a structure to it and we understand that structure pretty well. So, we don't necessarily need to get the entire genome sequences, but we typically interrogate something like between one to five million positions in the genome. So for each person in the study, we might get one to five million pieces and I'm sure you and probably most of your audience realize that if you want to detect a signal out of that number, you've already got a problem because you've encountered the multiple testing issues.
Jonathan Flint: 00:28:08
If I apply a standard of 5% significant threshold out of one million tests, I'm going to have to divide 0.05 by a million so I'm already down to 10 to the minus eight. And that means that I'm detecting a very weak signal with a very high threshold so that sample sizes go up and up in order to address both those issues. And we typically run sample sizes in the orders of tens, if not the hundreds of thousands. And there are now studies being published, which exceed one million subjects.
Jon Krohn: 00:28:43
Wow. And that makes a lot of sense to me. And so, that's a big part of this Depression Grand Challenge, right? That's part of why it needs to be a half-billion dollar project because each of the genes contributing to depression contributes such a small amount of the variants in whether somebody is depressed or not. And then, we have potentially tens of thousands of these genes interacting across millions of positions in the genome. And then yeah, these very, very small p-value thresholds because of this multiple testing problem and so yeah, so millions. Is there a target for the Depression Grand Challenge as to how many-
Jonathan Flint: 00:29:23
Well, when we go out and fundraise and trying to get people interested we say 100,000, but the reason we say 100,000 is because it's a big number really. I mean, we don't honestly know quite how many is going to be necessary. I suspect that may be an underestimate. It really depends upon the answer to the first question which is, how many diseases are we dealing with? And if we could identify a subgroup in that, which was more homogenous and would give us more of a signal, then we could get away with it. But we don't know the answer to that yet, but let's say a 100,000 is a reasonable guess.
Jon Krohn: 00:29:58
Struggling with broken pipelines, stale dashboards, missing data? You're not alone. Look no further than Monte Carlo, the leading end-to-end data observability platform. In the same way that New Relic and Datadog ensure reliable software and keep application downtime at bay, Monte Carlo solves the costly problem of data downtime. As detailed in episode number 499 with the firm's brilliant CEO Barr Moses, Monte Carlo monitors and alerts for data issues across your data warehouses, lakes, ETL and business intelligence, reducing data incidents by 90% or more. Start trusting your data with Monte Carlo today. Visit triple www.montecarlodata.com to learn more.
Jon Krohn: 00:30:48
And then, I suppose it also becomes more complex if you want to have more and more broad applicability across the planet. So, what's true for a European population may be different for an Asian population, for African population.
Jonathan Flint: 00:31:03
That's absolutely true and we've already begun to tackle this. As you probably know, I've been running large studies in China and we've had some success there. We've recently been funded to set up a large study in South Korea and some of my colleagues here are working both in South America and Africa collecting large samples for exactly that reason. And there's another reason also, which is that as you say, the variants that may predispose in one population are not necessarily quite the same as another. And that allows us to do some interesting comparisons between populations. So, if we think a particular variant is important in one population, we can test whether that might or might not be true by looking at its presence or absence in the second population. And that will give us better information as to whether it's causative or not. So, we can both exploit the differences between populations to help us, as well as use the ancestry differences to make genetic diagnosis in the different populations.
Jon Krohn: 00:32:01
Right. So, if it's the case that there's genes that are common between flies and people in learning, for example, then it could be the case that there are some common genes across different ethnicities in one species that are causative of depression.
Jonathan Flint: 00:32:22
Yeah.
Jon Krohn: 00:32:23
Cool. All right. So, other than going around telling people how big your project is and how much money it's going to need, what's your life like day to day as a professor at UCLA on this Depression Grant Challenge?
Jonathan Flint: 00:32:39
I would say about half of our time is spent fundraising and-
Jon Krohn: 00:32:44
Really, half of your time?
Jonathan Flint: 00:32:45
Yeah, pretty much so it's either grant writing or involved in philanthropy and whatever, we prostitute ourselves for anything really, I mean, we just want the money so we'll do whatever it takes.
Jon Krohn: 00:33:02
All right. And then, the other half?
Jonathan Flint: 00:33:04
The other half is mostly for me because although I'm a clinician, I don't treat patients here, is on the research, the teaching, the mentorship that goes with that.
Jon Krohn: 00:33:16
Research, teaching, mentorship. So, you have a lab, so the grant money comes in and then you have PhD students and postdocs.
Jonathan Flint: 00:33:30
Yeah.
Jon Krohn: 00:33:31
So, maybe you're directly mentoring postdocs, I guess, primarily.
Jonathan Flint: 00:33:35
Yeah, mostly. Yeah.
Jon Krohn: 00:33:37
Yeah. And then they have ...
Jonathan Flint: 00:33:39
Yeah. At the moment, because of this, we work with very large data sets. A lot of this is dry lab stuff, it's computational. And as you know, computational people tend to be not that social, they like to stay home in front of their computers. So, a lot of this is done remotely. So, if you ask where my lab is, it's sort of distributed. Some of the people aren't even in the city and we run collaborations. I work with people in different parts of this country and all around the world. So, the model has rather changed from just being the single investigator with these postdocs and PhD students working away in a laboratory to a much more distributed type of structure.
Jon Krohn: 00:34:21
Right. So, the computational part, obviously that's a big part of what you do and from even my time a decade ago, working on these problems with you, we had big dedicated servers, you were always getting the state of the art, you had access to super computing clusters for some of these problems. So obviously, there's that computational component and a big part of what your lab does is tackling these computational issues because of the kinds of problems you're describing, huge data sets for each individual. So, just talking about the scale for your project alone, you're talking about hundreds of thousands of people and for each of those people, you have millions of pieces of information about their genetics.
Jonathan Flint: 00:35:05
Yeah.
Jon Krohn: 00:35:05
Plus also information about environmental factors that they encountered, behavioral attributes, specifically related to depression, so huge amounts of data. And obviously, that ends up being a big part of what it means to be a geneticist today, for the most part, except for those people sucking up fruit flies in their plastic tubes. I can't remember how you phrased that. Is there still a lab component? Do some people of your PhD students or postdocs, are they involved in the collection of samples or is that contracted out? How does that work?
Jonathan Flint: 00:35:45
So, let's break that down a bit. We do run lab-based projects and that's really picking up on your question about if we found genes, what's the value and how can we understand them? Because we would like to understand the mechanisms and we need to think quite carefully about how to do that. And that ends up being a combination of molecular biology and some functional testing and so on. So yes, we have a small amount of people doing that. But fundamentally, I think that we need to make progress with this question of subtyping, this question of gene identification before we release, move it into the laboratory. So, most of what we're doing has been focused on the sample collection. And the sample collection because it's on a large scale has to be, I wouldn't say automated, but it's not contracted it out to another group but we work with hospitals.
Jonathan Flint: 00:36:40
I currently spend a fair amount of time working with our South Korean colleagues. We translated, obviously with their help, our assessments, we use a REDCap system for distributing and collecting the data on the subjects. We need to train the physicians so that they know how to get the best answers. And then, we have all of the downstream data analysis and QC that follows from that. And as an aside, one of the things that we have done is we've been recording the interviews of the depression. And the study that we carried out in China included about 10,000 people, it was all women. We chose women because the genetic effects on women are not the same as the genetic effects on men. And it's another example of heterogeneity so we wanted to make it as homogenous as possible.
Jonathan Flint: 00:37:37
And they're all between the ages of 40 and 60, really. So, we chose them and they all have recurrent major depression found in hospitals. And about 80% of those agreed to have their interview recorded and we edited those. We check, we have a group of editors who listen and if we think that they haven't made a good diagnosis, then we feed back to the interviewers and we may edit as a consequence data. And after that study was completed, we have these 8,000 recordings and I suddenly thought, "Well, why don't we use this as a phenotype?"
Jon Krohn: 00:38:10
Yeah, that was my next question.
Jonathan Flint: 00:38:11
We recognize people by their voices, it's a very individual feature. You can recognize when you pick up the phone and it's your mum, you tell within a few seconds. And we also know that you can recognize mood. We say, "Something wrong with you, Jon, are you feeling okay? You're sounding a little sad." So, what about us do we recognize in your voice that I say, "You sound a bit sad today." So, this is a problem that of course, really falls into the realm of the electronic engineers. And we set up a collaboration with the engineers here. It's a great thing to do in UCLA, very collaborative place. And I said, "Can you extract data from these recordings and tell me without knowing which are the depressed patients and which aren't?"
Jonathan Flint: 00:38:58
Now, my first surprise was that the vocal engineers here, the audio engineers, they take out hundreds of features so we gave them an hour and a half recording. And just from a couple of minutes, they're extracting 700 features. There's a very rich data set for them and they can do this of course, longitudinally so they've got a whole hour to go through. So, not only are they pulling this out as it were horizontally, but they can look across the whole interview and extract data. So then we thought, "Okay, so now we've got a very high dimensional data set out of phenotypic level, instead of just saying plus or minus depression, we've got 700 features and we've got 10 million data points on their genetics and we got this on 8,000 people."
Jonathan Flint: 00:39:39
So, we got 10,000 by five million by 800, this is an enormously complicated data set now so how can we use that in a more sophisticated fashion? Identify subgroups, can we use it to identify features, which would be predictive of depression. And it has the longer term goal that potentially, if this all works well, that we can just have a recording going on let's say, in the emergency room in the background or when a patient rings up a doctor, [inaudible 00:40:13] just automatically processing the recording and an alert comes up, "That patient's depressed, please go and speak to them." Maybe we'll be picking up before the subjects themselves know and I think that allows us to intervene earlier and potentially save lives.
Jon Krohn: 00:40:30
Wow, that is super interesting. I had no idea that that was going on, makes a lot of sense to make use of those data and brings me to my next question, which is what kinds of tools do you use or I guess, does your lab use to be handling these kinds of massive data sets? So, I know when I was doing my PhD with you, I mostly used R for everything, some MATLAB and now, I'm basically Python all the time. So yeah, you were using R all the time as well, that was your primary thing a decade ago.
Jonathan Flint: 00:41:02
Yeah. Well, I have to say I hate Python with a great hatred. Its memory is crap and you have to get 10 different versions of it because some modules in one version don't work in another, it's oh. But everyone uses it, Jon so I'm losing that battle, everyone's still using Python and R. And the other thing I'd really wish they'd move over to and I've had this conversation with a number of people is Julia, which has got much better memory handling capacity. It's much cleaner, runs fast, but unfortunately doesn't have all of those nice, easy libraries that you can just plug in. So, if you want to run some regression, you may even have to write the code yourself, Heaven forbid.
Jon Krohn: 00:41:47
Yeah, that is something ... I haven't dabbled in Julia at all but I did, near the end of last year, I did a series of episodes on the highest paying data tools and Julia emerged as something that you really should be learning if you want to get paid more as a data scientist because it's a relatively niche but the people who do use it command higher pay and it seems like these kinds of efficiency things that you described that are shortcomings for Python, yeah it seems like that is primarily why Julia was created if I remember correctly.
Jonathan Flint: 00:42:24
Yeah. I mean, it's dealing with those. I mean, I was amazed to discover how poor some of the memory demands are in Python that we were always having to fiddle to get extra requests on nodes and things to get it up and running, whereas Julia doesn't have those problems, it's been a much better software.
Jon Krohn: 00:42:46
Very cool. Well, so are there any particular R or Python or maybe even Julia packages that come up a lot in your work? I remember back when I was working, there was the Bioconductor package was a big thing in R [crosstalk 00:43:00]-
Jonathan Flint: 00:42:59
Yeah. So, that's really for a lot of the genomic analysis and so on. But to be honest, a lot of this is now just done with, we're interested in ... One of the groups that I work with does a lot of work on contrastive PCA and various data reduction techniques in order to pick up signals, so those approaches are becoming increasingly important. And I've been pressing our route to think more about machine learning analysis to deal with this, but we're not there yet by a long way. I mean, to give you a simple example of the sorts of problems we've got. So currently, we make a diagnosis of depression by talking to you. So, I sit you down and I say, "Are you feeling depressed and how long for?" And I ask you these questions. And then, I ask you about how that's been going on for the last two weeks. And if you break that barrier, then I give you a diagnosis, you meet criteria. Now, I have no idea what'll happen when you leave my clinic and I'm going to see you in two weeks. And I have no information from you apart from what you will give me when you return. And we want to break down two barriers, we want to break down this entirely subjective dependence on subjective information and we want to break down our dependence on getting information from you at a single time point.
Jonathan Flint: 00:44:31
And the inevitable tool, [inaudible 00:44:35] it's my phone, is to take information off your phone and we piloted this in a group of about three or 400 students. And they have been carrying around the Aware platform. It's a freely available data extraction tool, it queries sensors on your phone and has been sending us information about very simple things. Activity, GPS, location, phone screen time. Not surprisingly perhaps, phone screen time is a really good predictor of sleep. First thing anyone does in the morning is turn on their phone and last thing they do is put it by their bed and it goes to sleep.
Jon Krohn: 00:45:12
Oh, I thought you were about to say that screen time, in terms of how much time somebody spends is a great predictor of depression.
Jonathan Flint: 00:45:20
Well, that's what we've been investigating and you won't be surprised that indeed we detect a correlation between activity levels, so if people are not moving around, they're not using their phone so much and that does indeed predict their mood state. So, I was quite surprised how we could do that. But if you think about how you might analyze these data, so I've got data, this is a relatively small sample for us, a few thousand students, but taken over a long period of time, six weeks. So my question is, how do I recognize those students who are depressed? And most people, I think consider that problem as a contrastive one, so I have a group of students who are not depressed and I have a group of students who are depressed.
Jonathan Flint: 00:46:06
What are the features on their phone sensors that separate those two groups? And you will see something, it's not a strong signal. But clinically, that's actually not what I'm interested in, I'm interested in a change in an individual. So, we have longitudinal data, so my question here was, why don't we reframe this? And the first reframing was just as a Hidden Markov model. So, could you think about the change in state just like this model that is an HMM. And over time, then see whether you're getting any changes in state which were beyond what you would expect. So, we'd now divided up into a training set, look at about 70% of the people and then take that into the next 30% and see how good our predictions are. And accuracy at the moment is surprisingly high with about 80 to 90% accurately predict. But although that's good, clinically, that might not be anywhere near enough because I'm really only interested in those people who get so depressed that they might kill themselves at this stage. And I want 100% accuracy for that, I can't afford to miss a single person. So, we have a way to go with this but at least we're beginning to think about this I think in the right way, we have the sorts of data sets that will help us.
Jon Krohn: 00:47:27
That ties nicely back to that church crypt in your childhood.
Jonathan Flint: 00:47:30
Indeed, it does. Indeed.
Jon Krohn: 00:47:34
So, it sounds like there is a growing opportunity for data scientists, machine learning practitioners, probably software developers to be involved in genetics. So, you're collecting huge amounts of data but it sounds like there's still way more analysis that could be done, it sounded like you're getting started on a lot of these projects. So, how does somebody get involved with you, with a project like this? And actually, I'd like to tell a quick story for the audience, which I think you'll find interesting too, which is how I picked you as a PhD supervisor. So, I'd come from an undergrad where I'd done an undergrad in neuroscience and I'd done lots of computational stuff. I'd done writing statistical and programming scripts to be analyzing relatively large data sets, I had done for two years in my undergrad, I did an fMRI project.
Jon Krohn: 00:48:36
And so, I imagined that going off to Oxford and doing a PhD there, I would continue to do brain imaging work. And Oxford still has this brilliant program. So, the Wellcome Trust, which until the Bill and Melinda Gates Foundation was created was the largest medical charity in the world. It predominantly spends money in the UK, it comes out of a pharmaceutical fortune from Henry Wellcome so they fund these brilliant four-year PhD programs, which in the UK is somewhat unusual. So in the UK, you typically get PhD funding for three years, but these four-year programs give you a one-year master's beforehand, wherein you can have lots of lectures from lots of different faculty members and you can do research projects in several different labs to see what's a good fit for you.
Jon Krohn: 00:49:29
And so this is great, even for somebody like me, who came from a neuroscience background, but it's especially great for somebody who comes from a physics background or a medical background and they are getting started in neuroscience for the first time. So, there's this year to get acclimatized to the field, to try lots of different projects, have lectures from lots of different brilliant lecturers of different fields. And so, in my first term in Oxford, you did, I think just one lecture, but I loved it. I loved your style, the way that you presented information, the way that you were passionate about what you're doing and the way that you answered everyone's questions. And so, I didn't really care what you were studying. I was just like, "I want to work with that guy."
Jon Krohn: 00:50:20
Although I had done some genetics in my undergrad as well and so I knew a bit, but yeah, mostly just wanted to be working with you. And that's actually, that's something I've talked about on other episodes before, but my career has been guided by that basically. Every step of the way pretty much, my career decisions have been based on, "Wow, what an interesting person. I would love to be working with this person." So then, and you actually, you tried to get rid of me. So, I asked if I could have a conversation with you, if I could do a master's project, one of my term projects with you for a few months and you said, "Oh, you don't really want to do it with me. Look at these other interesting projects that other labs are doing, this would be far more interesting for you to do." And I said, "Nope, no, no, no, I really want to do this with you."
Jon Krohn: 00:51:08
And then, when it came time for me to choose at the end of that master's, so one of my big master's projects was a brain imaging project and the other one was with you. And I loved the other project, I think that would've been a grand PhD as well but yeah, really wanted to do that with you. And so, I can vouch for working with you as a wonderful experience and I gained a tremendous amount as an academic at that time and then through the rest of my career. You're quite a pragmatic leader and an outstanding communicator, not just in terms of lecturing but also in terms of communicating what you're looking for in somebody. And you put a lot of time into helping everyone out in the lab and succeeding at their project. So, I can highly recommend working with you, so how does somebody get involved in working with you on the Depression Grant Challenge?
Jonathan Flint: 00:52:06
Well, let me say two things, Jon. First thing, I want to thank you for the encomium and I'm very pleased that you used your coding skills that you learned in our laboratory to good effect. But just to reiterate what I said at the beginning, particularly for dealing with these psychiatric conditions that I mentioned. Anyone can help, anyone can get involved. And we'd love to have you in the lab and get you involved in various other ways as a scientist. But can I just also point out that this is something that anyone can help with. If you see someone who's depressed, please reach out to them. There is a belief that if someone is depressed and you ask them, "Have you thought of committing suicide?" You might precipitate that. It is not only not true, it'd actually could help people if you were to say that. And you may even save somebody's life.
Jon Krohn: 00:52:54
Wow.
Jonathan Flint: 00:52:55
So, listening as I've pointed out at the very beginning of our talk is important, just listening. And then, reaching out to those who are desperate. Don't hesitate. So, those are two very simple things that anyone can do. Now, if you want to be more proactive, if you want to come and help us work out what the causes are as a first step towards developing new treatment, send me an email. You can do stuff, we'll find something for you to do. I mean, this is such a big problem and we are so distributed across the planet that I think you're now realizing we're running programs in so many different countries. We want to do this everywhere. So, anyone can be involved at any level.
Jon Krohn: 00:53:38
And I imagine also at the donor level, if they were-
Jonathan Flint: 00:53:42
Well, funny you should mention that but yes, the one thing we're always short of, as you discovered when you asked me what I spend most of my time doing is unfortunately money. So yes, if anyone wants to contribute that way, but any form of contribution. Financial would be great but if it's just your time, your energy, that's also great too.
Jon Krohn: 00:54:01
Awesome, really appreciate that. Again, another pragmatic piece of advice. I didn't know that, for example, I didn't know that by asking somebody if they were contemplating suicide, I was in the camp that I thought that maybe that would be something-
Jonathan Flint: 00:54:17
Yeah, everyone thinks that. Yeah. No, don't hesitate. It's hard to say and don't tell yourself off if you don't manage it because I know what it feels like when you're sitting with somebody who's got their head in their hands and doesn't want to talk with you and reaching over them to say are you thinking of killing themselves seems like the worst thing to be doing, but it's not.
Jon Krohn: 00:54:37
Well, there you go. So, great piece of advice there. So, I would like to turn over now to some audience questions. So, I asked on LinkedIn and Twitter a week before we filmed to see if there were any particular questions that people had for an expert in psychiatric genetics, such as yourself. And some of them we've actually answered in the course of the discussion here. So, Mohamad Hussien, who's a data scientist asked what if depression has nothing to do with genes? We've already talked about that. We know from the twin studies that it certainly does and lots of other subsequent studies. But here's an interesting one. So, I think it follows on from his question of what if depression has nothing to do with genes, he asks, which cultures have the least depressed people? Is that something that you have insight into?
Jonathan Flint: 00:55:31
It's a very good question. We don't have a really good answer to that. And the reason we don't have a good answer to that is because the difficulty in making a diagnosis. But that said, for those countries where people have made the effort, a big effort to go out and interview people and try and make sure, it does appear that East Asian countries have less depression. I'll tell you one anecdote though just on this, which might also cast some light on some other issues. I was told this by a journalist. So, some years ago, no many years ago now, drug companies were not selling any antidepressants in Japan. And of course, they want to sell antidepressants, it's a good money earner for them, so they're very surprised, is it because there's no depression in Japan? That doesn't make much sense.
Jonathan Flint: 00:56:27
So, they commissioned some people to look into this and the result came back that it was really a question of how the question about mood was being framed. And that for the Japanese population going in and asking whether you were depressed wasn't working. So instead, they came up with a following. They had an advertising campaign and on the advertising campaign, a man turns up looking like he's suffering from Covid-19. His nose is streaming, he's coughing, he's looking a bit unwell. And the audio says, "You have a cold, we'll give you a remedy for your cold." And then, it cuts to a man with his head in his hands, looking very unhappy. "This person has a cold on the heart and the remedy for the cold on the heart is the antidepressant."
Jon Krohn: 00:57:16
Ah.
Jonathan Flint: 00:57:17
And very happily for the drug companies, sales of antidepressants rocketed.
Jon Krohn: 00:57:25
Right. So yeah, there is a big cultural thing here, where it's hard to know for sure if we're defining it in the same way.
Jonathan Flint: 00:57:30
Yeah.
Jon Krohn: 00:57:31
And yeah, that occurred to me. I didn't want to completely derail the conversation earlier but when you were talking about having your studies be used in Korea, in South America, so yeah, you have to train up people to be collecting the data properly, collecting saliva samples or whatever properly. But the hard part would be making sure that you're asking the questions in the same way when there's so much cultural influence on just what depression means. So yeah, I imagine it's one of the trickier bits.
Jonathan Flint: 00:58:07
Well, that's why we spend a lot of time training our interviewers and why we have a set of editors who listen in, why we record the interviews to check on this. There's a whole other science around how you acquire information from people. And you won't be surprised to know that the advertising agencies have most experience about this, how do you frame the questions and how do you collect the information? There's things like interviewer fatigue that after you've been giving the same interview for some months that you're not going to be listening so much and you'll rush through the questions or all sorts of little details that we need to be careful about dealing with.
Jon Krohn: 00:58:44
Right, very interesting. All right. So, another question here comes from Hank Yun, who is a data integrations engineer at Riskified and he's also a former student of mine on my deep learning course that I used to offer at the New York City Data Science Academy. And he would love to know if your research is more hypothesis-driven or data-driven or does it really depend on the situation?
Jonathan Flint: 00:59:10
So, we use the data in order to generate hypothesis and as you gathered from everything we've been talking about today at the moment, we're not really in the hypothesis generating stage, we have some general questions we want answered and the data will help us, I hope begin to formulate much clearer hypotheses.
Jon Krohn: 00:59:29
Cool. A great question from Hank and a really clear answer for me, Jonathan. Thank you. So, some of this we've already talked about. So, we have questions here from Serg Masís who is a data scientist and an author on interpretable machine learning and he was also recently a guest on the program. So, he was a guest on episode number 539, a really brilliant episode that I can highly recommend. So, Serg was interested in something that we already asked about earlier which is, what proportion is genetic and environmental? But he also asked how much of it is epigenetic, which is something that we haven't touched on at all in this episode, Jonathan. So, what is epigenetics and how is that relevant in depression?
Jonathan Flint: 01:00:18
So, epigenetics is the sort of thing that everyone resorts to when they can't get genetics to work and they throw out their hands and they say, "Oh, it must be epigenetics." But more seriously, epigenetics simply it's the study of gene regulation, how genes are turned, the transcription of the genes are turned on and off and how all that information wrapped up in your DNA is converted into something useful for the cell. So, that sort of effect is really what we will be interested in looking at once we know the primary questions. There's little point in looking at gene regulation until you know which genes to look at and we won't know which genes to look at until we've solved these questions about heterogeneity and underlying genetic causes.
Jon Krohn: 01:01:07
Outstanding answer, Jonathan, thank you very much. I learned a lot in that response as well. And Serg has a completely unrelated question, except that it's also related to depression which is, is there indeed a depression epidemic brought on by social media and mobile devices? And if so, what measures can be taken with technology to tackle that depression epidemic?
Jonathan Flint: 01:01:37
Well, there's certainly a depression epidemic being inflicted upon all of us now by the pandemic. There's increasing evidence for that unfortunately, due to the isolation and so on. I don't see very much evidence that social media can be blamed, whether it can be used to help, I don't know. I'm not a good person to ask about this, I have a deep hatred for social media.
Jon Krohn: 01:02:02
Come on, Jonathan, it's very trendy right now to be [crosstalk 01:02:04], we need your help here.
Jonathan Flint: 01:02:07
[crosstalk 01:02:07] I do use it so it's not like I don't look on Twitter for information and so on. But no, I don't know really, I don't really, that's a good question. I would need to think a bit more carefully about it.
Jon Krohn: 01:02:15
Yeah, no worries. It's not exactly a psychiatric genetics question, more of a general interest.
Jonathan Flint: 01:02:23
More of a general interest, yeah.
Jon Krohn: 01:02:25
But very nice to get your response on that. So Jonathan, a question that I rarely ask my guests, I save it for very special occasions and you are certainly a very special guest. I'd love to frame this time that we're at as truly unique. We have ever cheaper data storage, ever cheaper compute, ever more abundant sensors everywhere, these kinds of experiments you're describing where you can be having a phone in somebody's pocket and using that information to be trying to understand genetics. I mean, even just having genomic data, collecting genomic data, it's a very specific example of how data collection and storage is getting exponentially cheaper as years go on. On top of that, we have unparalleled intricate activity, you can be collaborating on these genetic data and behavioral data with labs all over the world. And those labs can be publishing papers or publishing code to GitHub and then everyone in your lab can make use of it. So, these kinds of innovations are having a big impact in every industry but in yours, in particular. So altogether, there's this unbelievable pace of innovation. So, what excites you about the future? What do you think could happen with technology in general or maybe in psychiatric genetics in particular?
Jonathan Flint: 01:03:55
I told you a little bit about this, I told you that we've been using phones to collect data on our students but we can definitely do better than that. And about a year ago, we signed a contract with Apple and Apple manufactures a watch. I thought watch is for telling the time and I thought maybe the Apple Watch was for telling the time plus answering your email and browsing your social media. But it turns out that Apple's main interest is actually in health and they market this as a way of monitoring your cardiovascular health. And they think that the biggest advance they've made is that your watch will remind you occasionally to stand up.
Jonathan Flint: 01:04:37
So, if there's anything that would improve your cardiovascular health, get a bit more exercise, just standing up occasionally is going to be useful. Now, they've sold cardiovascular disease, they want to take on psychiatric disease. And we've had a number of conversations with them, we went up to that big UFO building in San Jose and chatted with their executives about this and pitched to them the idea that we become their center for testing their technology and we've been doing that now for about every year. And we're collecting information from their watches and from a series of other sensors, which will give us, we hope the most rounded total view of how much information these sensors can give us that will actually allow us to predict something about mood.
Jon Krohn: 01:05:26
All right. So, that is a very specific example in the not too distant future, but what do you think beyond that? I mean, in decades, what could be possible? If we had millions of genomes sequenced and great behavioral data from the Apple brain implant or at least body implant that we have in the future that has lots of biochemical data being tracked in it, is there something for the next generations, might depression be treated in a completely different manner than it is today?
Jonathan Flint: 01:06:05
I will be very surprised if we don't make a radical advance along the way you've suggested. And I think there are probably two lines of research, which are going to give us these sorts of breakthroughs. One is the one I've just mentioned, which is collecting all of these data sets and I think in the end, we're going to work out that there are a few bits of useful information that we can get. And ideally, we would like for you to be told before things get bad, we really want to prevent disease rather than to have to cure it. And if there were simple, I mean, just going back to the Apple Watch idea that I can get you to stand up and prevent you getting cardiovascular disease. What type of interventions would our information givers that would say, "Do this, don't do that. What are good mental health behaviors which will prevent you getting depressed?"
Jonathan Flint: 01:07:01
And there's a lot of work in cognitive therapy around this, teaching you how to deal with situations. But they may turn out to be much simpler interventions based on your behavior and what we learn, which would help. What those are? I have no idea. But I suspect that will turn up in the next five to 10 years. And the second thing is that as we move forward with our ability to interrogate the entire genomes and to go back to some of the conversations we had earlier, we are learning so much more about brain function. I mean, the advances in neuroscience are staggering. Our ability now to interrogate individual neurons, their contents, we can pull out, suck out from that little cell, all the little things that it's got inside it, we can categorize them, catalog them, see how they match up with what other cells are doing.
Jonathan Flint: 01:07:51
And increasingly, we'd able to do that and interrogate the function at the same time. We talked a little bit about optogenetics and the access we're getting into mammalian brains to see what they do. So within 10 years or so, that information should be giving us much clearer idea of what happens to a brain when it's depressed. And I suspect we'll come up with some great surprises. The thing that maybe you and your listeners don't know and it's always a bit of a surprise this, is that although we have these wonderful technologies and we know so much more than we ever did about what brains consist of, our ideas, our hypotheses about how brains found are really limited at the moment. There are very few theories. And if you want some fun, go into a bar where there's a lot of neuroscientists and just ask them what a memory is, what is a memory? How's it laid down?
Jonathan Flint: 01:08:51
And even simple things like, is it digital or is it analog and where is it in the brain? There'll be a lot of hand waving and there'll be some arguments and so on, but we definitely need breakthroughs just in generating hypothesis about how brains work. And that's another issue for your listeners. If you want to get involved, start reading these papers because we do need people who think out of the box to come in and really understand what type of questions we should be answering. And then, I think one other thing that's just worth bearing in mind, this is another conversation I've had here. As you know, artificial intelligence is pretty good now at allowing us to speak to computers. And does this mean that if I write let's say, a grant proposal in which I say, "I want to understand how brains interpret sound because I'm interested in some application of that."
Jonathan Flint: 01:09:52
Then my grant will get turned down because the reviewing body will say, "Well, we don't need to know that because we have AI and AI has solved this problem with how brains process sound because we can just do it on a computer." To what extent in our ability to make predictions based on large data sets? Are we missing, avoiding perhaps necessarily, the need to come up with hypothesis-driven science just because things become so practical? And if I could apply some big AI machinery to all of the data we've collected and just use it to predict who and who is not depressed, bypassing any indication of the biology don't need any genetics, just machine learning, brute force approaches so that I can intervene and start giving them treatment. Is that going to be enough? And I think that's a question we don't really address perhaps for the seriousness that it deserves.
Jon Krohn: 01:10:48
Yeah, that makes a lot of sense to me. It is interesting how a lot of interventions we don't understand though, like anesthetics, I don't think, we don't know how anesthetics work, how people can temporarily not be conscious and then be conscious again and antidepressants as well, right?
Jonathan Flint: 01:11:05
Yeah. We don't know any psychiatric drug works.
Jon Krohn: 01:11:08
Right.
Jonathan Flint: 01:11:08
I mean, we don't even know why alcohol makes us relaxed and happy. Substances, ethanol, it's got about six constituents, why don't we understand even that?
Jon Krohn: 01:11:23
Right. Yeah. I mean, we don't understand how consciousness arises.
Jonathan Flint: 01:11:28
That's an even bigger question so ...
Jon Krohn: 01:11:33
It seems like we basically don't know anything, but we can collect a whole bunch of data so we're going to keep doing it. We're going to genotype more people-
Jonathan Flint: 01:11:39
We're data scientists, we're computer scientists and data scientists, we will just collect data, that's true.
Jon Krohn: 01:11:45
Yeah. All right. And then, I'd like this idea you're talking about looking to the future and what an Apple Watch or what technology could be doing. And so, maybe in addition to Apple watches in the future telling you to stand up, they'll also ask you, "Are you contemplating killing yourself?"
Jonathan Flint: 01:12:07
Maybe that'll happen, who knows? I hope it's in a slightly more sophisticated fashion than that.
Jon Krohn: 01:12:16
All right. And then Jonathan, we end all of our episodes with a book recommendation. Other than your own book, do you happen to have one for us?
Jonathan Flint: 01:12:24
Well, I have the second edition as well as you know, so I think that is at least two book recommendations.
Jon Krohn: 01:12:28
Once you get through the first then try the second.
Jonathan Flint: 01:12:31
All right. Well, the book that I have been reading a lot and keep going back to is by Scott Aaronson, it's called "Quantum", don't let me get the title wrong, "Quantum Computing since Democritus," or something like that. It says on the back and it's true, "This book will make you fall about laughing."
Jon Krohn: 01:12:52
Okay, great. That sounds like a fun one. Actually, I need something funny to read. I often want something fun to read. I spend my day reading quite serious things. And yeah, often struggle to find something funny. Read a bunch of Douglas Adams for a while and that's pretty funny, but yeah I don't know, I want something else.
Jonathan Flint: 01:13:13
Okay. Get a Scott, he'll tell you something will make you laugh.
Jon Krohn: 01:13:17
Nice. All right. And then Jonathan, how should we get in touch with you? You mentioned earlier that we could just email you or something-
Jonathan Flint: 01:13:24
Email's best, yeah. jflint@mednet.ucla.edu or you can just look me up on the internet, it's pretty accessible.
Jon Krohn: 01:13:32
Yeah, we'll be sure to have your website as well as your email address in the show notes. All right. Jonathan, it's been absolutely wonderful to catch up with you today. Thank you so much for joining me on the SuperDataScience Podcast and maybe we can have you on again in the future to give us an update on [inaudible 01:13:50].
Jonathan Flint: 01:13:50
It's been a great pleasure, Jon, thank you very much. I look forward to our next meeting.
Jon Krohn: 01:14:00
Well, can you see why I chose Professor Flint as my doctoral supervisor? My goodness, it's remarkable that somebody so extraordinarily industrious and intelligent is also capable of such deep empathy and clear communication. He's somebody who's changing the world dramatically and I'm honored that he was willing to spend time with us on the show. In today's episode, Jonathan enlightened us on how studies of twins indicate that about 35% of the susceptibility to psychiatric disorders like depression is genetic. How the single disease that we call depression today is likely to be many different diseases, each with its own distinct genetic basis. How computational techniques like statistics and machine learning are critical to identifying causal patterns within the vast troves of genomic data collected today and how this trend will only increase in the years to come.
Jon Krohn: 01:14:52
How the programming language Julia has been markedly more memory efficient than the more widely used Python language in his experience, how contrastive principal component analysis and Hidden Markov models have been particularly useful techniques for him for analyzing genetic data and we could even save a life by asking whether someone is thinking of taking their own life. Now, that's practical advice. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Jonathan's Twitter profile, as well as my own social media profiles at superdatascience.com/547. That's superdatascience.com/547.
Jon Krohn: 01:15:33
If you'd like to ask questions of future guests on the show like several audience members did of Professor Flint during today's episode, then consider following me on LinkedIn or Twitter because that's where I post who upcoming guests are and ask you for your thoughtful inquiries. All right. Thank you to Ivana, Mario, Jaime, JP and Kirill on the SuperDataScience team for managing and producing another deeply stimulating episode for us today. Keep on rocking it out there folks and I'm looking forward to enjoying another round of the SuperDataScience Podcast with you very soon.
Show all
arrow_downward