BI 218 Chris Rozell: Brain Stimulation and AI for Mental Disorders

[00:00:03] Speaker A: You know, increasingly, with the way brain computer interfaces are making their way into popular culture and popular language, we can even start describing it as a type of brain computer interface. But for people with psychiatric conditions, it's bigger than any one of us. It's bigger than any one thing that we can hold in our own expertise. And so bringing together scientists of different stripes, engineers of different stripes, clinicians of different stripes, stripes, the lived experience, voices like the magic for me happens where all that comes together. Take a piece of brain data, put it through the encoder, get into the latent space. If we then move along that one special dimension, what changes about the brain data when we reconstruct it through the decoder is something that will change that classifier's mind about whether that brain was sick or well. [00:01:05] Speaker B: This is Brain Inspired, Powered by the transmitter. Hi everybody, I am Paul. We, as in humanity, we are in an exciting time in the cross fertilization of the neurotechnology industry and the cognitive sciences. My guest today is Chris Rozelle, who sits in that space that connects neurotech and brain research. Chris runs the Structured Information for Precision Neuroengineering lab at Georgia Tech University and he was just named the inaugural director of Georgia Tech's Institute for Neuroscience, Neurotechnology and Society. I think this is the first time on Brain Inspired we've discussed stimulating brains to treat mental disorders. I think today we talk about Chris's work establishing a biomarker from brain recordings of patients with treatment resistant depression, a specific form of depression that we discuss. So these are patients who have deep brain stimulation electrodes implanted in an effort to treat their depression. Chris and his team used that stimulation in conjunction with brain recordings and machine learning tools to predict how effective the treatment will be, under what circumstances, and so on to help psychiatrists better treat their patients. So we'll get into all the details and surrounding issues of that study and that work. Toward the end, we also talk about Chris's unique background and path and approach and why he thinks that interdisciplinary research is so important. He is one of the most genuinely well intentioned people I've met and I hope you're inspired by his research and his story. Check out the show notes where I link to the paper that we discuss and other relevant information that we discuss. Thank you to my Patreon supporters. If you value this podcast, go to patreon.com braininspired good things await you if you support this podcast through Patreon. And thanks as always to the transmitter for their continued support as well, enjoy. [00:03:17] Speaker A: Chris. [00:03:20] Speaker B: Chris. One of the many things, a handful at least, maybe not many. A handful of things that I'm jealous of you about is that. And we'll get to your new position in a minute. But when you're sitting around the dinner table with family and friends and your cousin asks you what you do, you probably have a much more straightforward answer than I do. I get frustrated because I flounder trying to describe what I do. But you can kind of easily say, how's this? I stimulate brains to treat people with depression, mental disorders. That's a very, very quick, poor overview. But it's something like that, right? [00:03:59] Speaker A: Yeah, something like that. And I'll say that wasn't always the case for me. You know, my career has evolved over time and it's been very gratifying to be able to work more recently on things that have a direct health implication. And it does make the description easier. We try to understand the circuits in the brain that go wrong when someone has an intractable psychiatric condition, like a treatment resistant depression. And then we're engineers, we want to understand how we can intervene in those circuits. And in our case, we do it often with things like deep brain stimulation, which is like a pacemaker type implant for the brain. And so it does lend itself to something that is very easy for the layperson to understand. Increasingly, with the way brain computer interfaces are making their way into popular culture and popular language, we can even start describing it as a type of brain computer interface. But for people with psychiatric conditions, rather than for people with a paralysis or something like that that you might be more accustomed to seeing. So it does take people by surprise sometimes because they don't know that it's a thing. But it is a nice succinct description and it is of course, very gratifying and a privilege to be able to do, to do work that, that has impact that people can appreciate. [00:05:21] Speaker B: Yeah, okay. [00:05:23] Speaker A: Yeah. [00:05:23] Speaker B: I will strive to obtain a position more like yours. Maybe I need to switch gears just so I can explain it more, more easily to people, what I do. [00:05:31] Speaker A: Right. Well, I started in theoretical neuroscience and it was a, it was, it was very difficult to explain to people, you know, what you were doing and also what the value. [00:05:46] Speaker B: Yeah, I know. [00:05:47] Speaker A: Well, the value. [00:05:48] Speaker B: Well, it'll pay off in 50 years, I promise. That's the sort of promise of that kind of research, right? [00:05:54] Speaker A: Yeah. And it's difficult because as a scientist, I believe so much in the value of it. But when you're talking to non scientists and into laypeople, Theoretical neuroscience is difficult to describe what the value might be, especially if it's not in the context of a health related condition. In the past you might have said something that tied it to we have to understand these things theoretically so we can build better AI systems. And I think with all the recent advances in AI and of course some of the kind of realization in the public consciousness about what an AI future might mean, I'm not sure if that would be as widely appreciated as it might have been, you know, 20 years ago when AI felt more like a distant future than it does right now. [00:06:48] Speaker B: Yeah, I hadn't thought about that. We will discuss a little bit of AI because you used AI in one of your recent research efforts that we're going to talk about that you also use deep brain stimulation to treat people with depression. This is a pretty major and special study and I think it's received a lot of deserved attention. Did this paper get a lot of attention? [00:07:13] Speaker A: It did. Which again, we're very grateful for the way it was able to drive some conversation both in scientific circles and in the public circles about the value of interdisciplinary research that went kind of well beyond the individual study. But it did and I think the reason for that and we can get into as much technical detail as you want. It was one of the first times that we've been able to record from deep in the brain while someone is undergoing, in this case, deep brain stimulation for treatment resistant depression. We're able to see something, even though it's still very limited about what's happening in the activity in those brain circuits. And we're able to measure something objective that's changing. And in psychiatric disorders where a lot of the assessment is done subjectively and that can be. [00:08:04] Speaker B: What do you mean subjectively? A clinician asks questions, a patient answers those questions, and then they infer a depressive or non depressive state from it. [00:08:13] Speaker A: That's exactly what I mean. It's through the judgment of the, in this case, psychiatrist observing the patient, asking them questions, surveys that either the clinician is filling out based on those observations or that the patient is filling out themselves. And they're, they're very non specific instruments in many cases, right? The, you know, if you have a condition like a cardiovascular condition, we can measure your blood pressure, we can measure your lipid counts, right. If you have diabetes, we can measure your blood sugar, right. We have data we can collect. If you have a movement disorder like Parkinson's disease, we can measure things like the tapping of your fingers and the speed of that or how long it takes you to walk across the room. We can measure something. And psychiatry is one of the few branches of medicine where we don't really have good ways to measure data directly. And that's what I mean when I say objective, that we can provide, you know, really concrete information about either a diagnosis or, in our case, we were trying to measure a recovery response and these subjective instruments. And this is, no, I don't want this to sound pejorative to our clinical colleagues because I think the psychiatrists that we work with are some of the bravest people I know entering into these spaces and are able to do incredible things out of the judgment that they've, that they've developed over years and decades. But I think they would also be the first to tell you that it's very difficult, especially when the patients are having a difficult time self reporting what's going on in the moment, distinguishing, just to be very specific, in our case, very difficult to distinguish what depression relapse might feel like as distinct from, say, an emerging anxiety, right. That can even be arising because the depression is remitting, right? And you're feeling new things that you haven't felt for a long time. You're maybe contemplating a new life and that's worrying. Am I still going to feel lonely even though my depression has lifted? Right. And in those moments, because of the dysregulation of the body and these emotional systems, those things can feel very similar. And the patients will tell you this, they'll say, you know, it's very difficult. The doctor keeps asking me, is this anxiety or is this depression? And they'll say, in the moment, I can't tell, right. Sometimes with the benefit of hindsight, they're able to go back and say, oh, that was different. That felt different, you know, in situation A from situation B, right? But in the moment, clinical decisions have to be made in the moment. They don't get to be made with the benefit of hindsight. And so one of the things we've been trying to do with these experimental investigational therapies is to think, how are they going to be managed by the clinical teams? Because the therapy isn't going to be adopted if the clinical teams aren't confident in their ability to management. And so we're trying to provide some objective information to inform their clinical decision making. So to combine together with all of their rich experience and knowledge, just as another piece of concrete information to help them distinguish what they're observing and then make decisions about, in this case, whether a stimulator needs adjusting or perhaps there needs to be an adjunctive therapy of medication change, psychotherapy, social support, something else going on. These are clinical definitions right now. Even the scope of something like depression. You're defined through these very broad, you know, kind of categorical and subjective statements. You know, there, there are almost certainly subcategories of things going on. And I think it's becoming increasingly, you know, a place of hopefulness that we can resolve specific deficits and specific circuits that are leading to the specific things that are going on. But we're not yet at the point that we can measure, you know, say, behavior and different clinical things, you know, objectively enough and in enough resolution that we can start to define these subtypes and then look at the specific circuit deficits that might underlie them. I think that's a, that's a hope that we all have for the future, that we can get to a place where we can do that. That's not where we're at right now. But I think one of the reasons this paper generated some interest was it was, it was kind of a first volley over the net saying, I think that there's hope here. I think we can measure something from the brain and we can tie it to something that's clinically meaningful and difficult to assess in the moment right now, which in this case is the patient's recovery status from their core depression as they're receiving this type of stimulation therapy. [00:13:19] Speaker B: Well, yeah, and in this case also, the measurements are somewhat sparse. Right. I mean, you're taking kind of a low, not low quality, but very few measurements for what is a complex mental phenomenon in depression, especially treatment resistant depression. But let me back up, because you were talking about the role of, the ongoing role of the clinician. And one way one could read this paper is, oh, no, clinicians are going to be. Psychiatrists are going to be replaced because here's this biomarker that can predict earlier than, you know, clinic, other clinical measures. [00:14:03] Speaker A: And. [00:14:06] Speaker B: So we can just replace psychiatrists. But you're careful in the paper also to say, like, this is, like this is just another tool in their toolbox, essentially. [00:14:13] Speaker A: Yeah, I don't personally see psychiatrists going away any anytime soon. We have a huge need, a huge need in society. [00:14:23] Speaker B: Someone's got to prescribe the drugs. [00:14:25] Speaker A: Well, and even in the context of this therapy, you know, I. The work that we're doing and the clinical collaborators that we work with, and primarily I'll mention here that Helen Mayberg is my clinical partner in this and really a pioneer in the idea that depression could be tied to a circuit deficit that could be modulated with things like dbs. She's a pioneer in every sense of the word. And did these surgeries for the first time over 20 years ago when DBS was first. Shortly after it was first approved for Parkinson's. I think their first surgeries were in 2003 for treatment resistant depression. Yes. [00:15:08] Speaker B: Oh, see, I'm so naive. Like, I did not realize how prevalent this kind of. That dbs. Well, I guess they all have. They're different. I associate DBS with Parkinson's because that's sort of the famous story, but I didn't realize that the treatment of. Of this kind of depression, that DBS has been used for that for so long. [00:15:30] Speaker A: Yeah. Investigational use. [00:15:32] Speaker B: Yeah. Okay. [00:15:33] Speaker A: So for approved indications, you have movement disorders, you know, so Parkinson's, essential tremor, dystonia. You do have some other indications, including epilepsy. And under a humanitarian exemption, it is actually approved for ocd, although it's not commonly done because many insurance carriers won't pay for it. [00:15:56] Speaker B: What does a humanitarian exception mean if it's just very severe OCD when you have. [00:16:03] Speaker A: And I'm certainly not a regulatory expert, but there are approval pathways when there are really a paucity of things that can effectively treat a disorder. And I think it's. Especially when it's a relatively small number of cases. So OCD being less common than, say, something like a depression or a movement disorder or something like that. But again, I'm not a regulatory expert, so that's my understanding of it. So depression was first tried, like I said, by Dr. Mayberg and her team in Toronto at the time. Andreas Lozano was part of that team, so over 20 years ago now. And there are a number of sites that have studied it in different ways using some different targets, Although the target that Dr. Mayberg's work has led to is probably the most common target. There are others that other groups have tried and other strategies, including personalizing the target for individuals, including doing closed loop stimulation adjustment. [00:17:11] Speaker B: Can you describe that real quick? Just what closed loop is versus open loop? Just for the. [00:17:15] Speaker A: So closed loop would be you're recording a signal, much like we talked about in this case, a local field potential signal. So an aggregator signal of the. Of the neural activity in the vicinity of the electrode. And in a closed loop system, there would be you kind of real time adjustments to the stimulation based on what was being recorded. You know, essentially when people use this word, they mean it in a way where there is no human intervention. You know, so think about something like cruise control or autopilot or the thermostat in your home that's measuring. You set up what want it to measure and how you want it to respond. But the system is going to automatically adjust the, the input, you know, whether that's the accelerator pedal or the, the furnace in your home or something, or the dbs. Right. So you're just the stimulation level in. I think most people would consider what we're doing open loop in the sense that that signal is not being used to directly drive the, the, the simulation changes in some sense. I think you could still say it's closed loop, but with human. [00:18:29] Speaker B: Yeah. [00:18:29] Speaker A: And on a slower timescale, I mean, we're, you know, we're making changes on, you know, on timescales that look like weeks. Right, right. Weeks to months. Because, you know, in, in our view of what's happening, it's a slow disease, it resolves slowly. And so those are the. That's the timescale that we're making adjustments. And when I say we, you know, I'm talking about the royal we, in a sense, of the whole clinical team, because there's a, you know, a psychiatrist that is managing the patients and making those decisions. We're providing your data to try and help support those decisions, or that is the goal, I should say. You know, the paper that you're referring to showed that there was information in that data that we should be able to use to inform. And that's exactly what we're working at right now, is how does one actually move to a full decision support system where you can inform in a way that is helpful to the clinic? So we've yet to show that you can actually be helpful to the clinicians by informing them through this data. [00:19:35] Speaker B: So I guess the main thing the paper showed then in that regard is just that you can predict recovery rates based on these LFP signals, which are fed through some machine learning and adjustments of the deep brain stimulation parameters. You can actually better predict the recovery than other markers that are known. How would you phrase that? [00:20:01] Speaker A: Yeah. In this case, what I would say we showed was we showed good accuracy in predicting week to week, whether the patient was still sick or had entered stable recovery, which you can only define retrospectively. Right. So for us, stable recovery means that their clinical scores of the survey instruments that we were talking to have dropped a sufficient amount. And there's a gold standard definition for what that means. And the stable part is for us, effectively, they stay there, they don't rise back up again within the six months of our trial. Accepting individual weeks. Right. Because individual weeks, these survey instruments can pop around. Right. The scores can be quite high. One week, and you talk to the clinical team and say, why wasn't an adjustment made there? You know, you were using your decision, your own insight, your scores got worse. You didn't make an adjustment. Why is that? And they'll go look in their notes and say, oh, that was a week where maybe there was a letter from the IRS that they were getting audited and owed money. They were really stressed. But we could tell in that case that it wasn't their depression returning. It was just a stressful event. Right. And so individual weeks can happen like that. [00:21:26] Speaker B: Is that one reason why you would not maybe want closed loop? [00:21:31] Speaker A: Perhaps? And I'll say there are groups exploring closed loop, and I'm very interested to see what they come up with. Just in our hands and in our approach to this. That's how we've approached it. But I would say that's one potential concern is are you adjusting it more than is necessary, and does that have any detrimental. Detrimental effect on recovery? I don't know. And I'm very excited to see what those groups are able to show with their approaches. This is a complicated disorder, and we need different shots on goal. And so we're all kind of taking different shots, and we'll see what ends up being useful or not. [00:22:11] Speaker B: Well, you're from Michigan, is that right? [00:22:13] Speaker A: Michigan. [00:22:13] Speaker B: A lot of hockey there. A lot of hockey with the shots on golf. [00:22:16] Speaker A: Exactly. I'm giving away my. My Michigan roots by using a hockey. [00:22:21] Speaker B: Yeah. One of the reasons why I asked you about that closed loop, you know, whether you might not want to do that is because one of the things that you and your team found is that there's this sort of initial change in the LFP signals shortly after the stimulation that is kind of counterintuitive to what you would maybe want to see. And then it takes a little time to level out and become. And be in register with the predictive recovery. Is that one. Is that a correct way of framing that? [00:22:56] Speaker A: Yeah. So that's a good example. The recovery of the patients in this trial follows kind of a sequence that takes time, takes weeks to months. One of the challenges is it's different for every person. [00:23:08] Speaker B: Yeah. [00:23:08] Speaker A: And so some are quite fast responders, some are, you know, are much slower, take many months. And of course, some don't respond to the therapy like any therapy. And so the heterogeneous trajectories. Right. The Individuality of how they're going to respond is one of the challenges. And we see it happen in sequences, both in the symptoms, but also in the electrophysiology. So in this case, I think what you're referring to is the signal that we're looking for is a combination of different oscillations. So you may have heard of theta, delta, alpha, beta, gamma band oscillations. That just means kind of frequencies or oscillations at specific frequency ranges. [00:23:57] Speaker B: How much of each of the frequencies is sort of dominating the signals at each time? [00:24:01] Speaker A: Exactly. So think of it like a symphony with different instruments, all different registers. So which ones are happening at a time, or in our case, which ones are changing over months. So in this case, eventually over the six months of our trial, we saw that there were some specific frequency bands that had to increase in order to indicate that someone had entered stable recovery relative to where they were when they started. And that's all well and good, but one of the potential, kind of interesting parts of this is that we knew from previous data testing when stimulation was first applied in the operating room, that one of these frequency bands, in this case beta band activity, went in the other direction. When you first applied stimulation, it went down. In fact, it didn't go up. And that was actually predictive of how that patient was going to be feeling in a couple of weeks. So the more it went down, essentially the better they were going to be feeling. [00:25:04] Speaker B: But also. But also it would go up with. With their feeling better. Right. [00:25:09] Speaker A: Is that eventually. Eventually, in our study, in the previous interoperative studies, we just knew that decreasing that frequency was going to be good later on. [00:25:20] Speaker B: Yeah. [00:25:20] Speaker A: Right. So if you wouldn't have measured over six months, if you would have just taken that first bout to stimulation, oh. [00:25:28] Speaker B: You'D think, let's depress beta and that'll cure them. [00:25:31] Speaker A: You would have to go in the wrong direction. Yeah. [00:25:34] Speaker B: Or keep it lowered. I shouldn't say depressed. [00:25:36] Speaker A: Yeah, yeah, yeah. Inhibit, inhibit. Right. And it was really, you know, the advances in our ability to record from the brain, in this case by, you know, our commercial partner, Medtronic, because they were interested in closed loop stimulation for Parkinson's disease, they designed a device that lets you record chronically over time, the local field potentials. Right. So now that we have that ability and we were able to, you know, approach something like the Brain Initiative from NIH and say there's been this technological advance, Right. We can finally do this stimulation, we can record, we can see what's happening over time. And we can try and make the sort of advances that are needed to take this investigational therapy and really make it scalable and accessible to the broader population that needs it. By measuring longitudinally, we could see that actually what we needed was an increase in this beta band activity for someone to be in stable recovery. Now when we went back and looked in the chronic recordings in these patients, there was also this inhibition of beta band activity in the first month when we turned it on, that eventually over a few weeks returns to baseline and then starts to facilitate as patients enter stable recovery. I think the lesson here is that it was really important for us to collect this data longitudinally over six months and see. And that technical innovation was really critical because otherwise we would have been in some sense chasing our tail a little bit by taking the short acting effect of the stimulation. And I think that's from a clinical point of view, that was a very important observation. But I think it's also scientifically really interesting, tells you that something that we might have suspected, given how long the therapy takes to work for a depression patient, that it's not just the, the impact of the stimulation directly that's affecting things like you often see in Parkinson's. Right. You change the stimulation symptom changes right away. [00:27:45] Speaker B: You turn on the stimulation and the tremors stop. [00:27:48] Speaker A: Yeah. Within, you know, minutes, effectively they can tell something about how that stimulation change has, has helped or not. That's not the case for depression. So one of the reasons we need these biomarkers to help them know if the settings are right. But it also tells you, or is at least a very strong clue that part of the recovery is some type of adaptation in the circuit. There's some type of plasticity going on. It's not just the direct effect of the stimulation. Right. Because there's something, it's nonlinear. Right. There's something happening in the opposite direction from when you first turn the stimulation on. So it's a powerful clue that there's something adapting in the circuit that's part of the recovery. And that's one of our challenges to figure out exactly what that means and whether that can be helpful in deep brain stimulation, whether that can give us clues to other methods of treating depression. [00:28:45] Speaker B: Right. I mean, one thing that springs to mind is there's a similarity, at least with modern use of psychotropics. Right. To. So one of the interpretations of this sort of longer term initial thing that then longer term kind of goes into a different regime is that you're sort of resetting the system. If you Think of your brain activity as like a surface with a bunch of sort of wells in it. Then this is a complete analogy. But depression, let's say you can kind of get stuck in a well and then by stimulating, you can think of it as sort of flattening out the wells and letting the brain settle back into a healthier state where you're in a better well for, for lack of better analogy. And you know, things like, well, psychotropic drugs are thought to work somewhat in the same way. I don't know. How do you think about that? [00:29:43] Speaker A: Yeah, I, and I have to say that I, I am certainly not an expert in psychotropic drugs, so I'm not going to speak too directly to that. But I do think that that matches our conception of what's happening with, with deep brain stimulation, that there is an acute effect, a short term effect, a fast acting effect that is a type of kicking the system out of a bad state. And we can see that happen sometimes with just a few minutes of stimulation in the operating room. [00:30:09] Speaker B: But this is in a very circumscribed brain area. You think of, if you take acid or something, it's like, oh, your whole brain is going berserk. Right. But this is like a very small brain area that you're. [00:30:21] Speaker A: It is, but this is a network disorder. And the targeting that Helen Mayberg and Ki Sung Choi and Patricia Riva Posay and others have very carefully done through imaging over the years is really focused on hitting kind of a highway junction. So it's targeted through, looking for an intersection of white matter bundles, four of them specifically, that reach through different major systems in the brain that we know are implicated in depression. So yes, we're targeting one specific area, but that area is not like a gray matter target like you might think of in Parkinson's. It's a white matter target that reaches into multiple major networks in the brain. It's really a super highway hub for many of the functional networks that you hear about. Default mode network, salience network, you know, the attention networks. It's, it's a, it's a hub where many of these networks have some intersection point. [00:31:24] Speaker B: Well, so I mean, then you got to be cautious because there's lots of stuff that could go wrong as well. Right. You could be resetting other mental phenomena in, in negative ways, perhaps. [00:31:36] Speaker A: Yeah, yeah. We don't see direct negative side effects of the stimulation. Sure, yeah, right. Which is obviously a concern that' monitored for. But I mean, you're affecting a network. And so, you know, certainly there you know, there are going to be many changes, we hope most of them, you know, the right changes that move people toward a recovery. [00:31:59] Speaker B: I wonder like what the. You probably, you might know the answer to this already because I'm not sure what, what's what your, how you followed up or planning to follow up. But, but I can see a scenario where there's like less default mode. So default mode network is often associated with sort of an internal monologue, your internal state and thinking and daydreaming and stuff. And I wonder because I'm not an expert on these things anyway, so this is, I'm talking out of hand here, but you know, you can sort of perseverate in those states when you're in a depressive state. And I wonder if it's associated with less default mode network essentially. But yeah, yeah. [00:32:37] Speaker A: So I'm certainly not an imaging expert either. But it would also be my understanding that you see in some sense an overrepresentation of things like default mode network, which is often kind of characterized by internally driven thought in the clinical context. You might call it kind of rumination. Rumination, Right. And so both an over activation and maybe some degree of over connectivity. But again, I'm not an imaging expert but I believe that that's been studied quite a lot with different types of therapies, both, both pharmacologic and you know, TMS and DBS and things like that. [00:33:17] Speaker B: But what you've, but what you saw in your study is that the, the white matter, the health of the white matter is also predictive since you know, you're talking about the superhighway of the, the connective cables essentially of between neurons. So, so the health of the white matter was actually very predictive of the depressive results. Right? [00:33:40] Speaker A: That's right. So just back to your previous statement about short and long term effects. We do see kind of short term effects, but then also these kind of longer term symptom resolution. And we have this notion that there's something kind of adaptive happening. We don't know yet what it is. But of course an intriguing sort of clue is what you're mentioning, which is the imaging team on that study was able to look at those patients and, and certainly there's been a lot of work trying to concord imaging findings with symptoms. This is a very small study. But now that we had an objective marker and not a more subjective symptom based description, the imaging team was able to look at the objective brain recovery and say, okay, this person responded very quickly. This person responded over Many, many months. You know what's different in the, in the brain imaging at baseline, because in these patients we weren't able to put them back in the scanner once they were implanted. We can do that now because the current commercial devices are Mr. Compatible, at least to some degree. But at the time we couldn't. So this was all before they had surgery. And I think one of the incredible things that the imaging team was able to find is that there was a correlation with, I'll say, structural abnormalities in the brain and it's imaging. So these are in some sense indirect measures, but it was most consistent with something that looks like demyelination. So some focal areas of myelin abnormality in specific places. And then using something like a functional connectivity, which is more of an FMRI based technique where you're looking at correlations between different brain areas. And looking at our stimulation site and other sites, there is a deficit in functional connectivity that spanned across that abnormality in the white matter structure. We don't know how that got there. Of course, we don't have that type of imaging data long term on patients or any other notion of what's going on. But it was a very compelling finding to me. One of the most exciting findings in the paper was this concordance. And it did also correlate with essentially the number of depressive episodes, so how long a patient had been sick before they enrolled in the trial. We don't know certainly the causal connection there, whether there is a causal connection, but there's something to look at there about whether a degree of illness either characterized by the history or how long it takes someone to get better, whether that is causally connected to these structural and functional deficits in the white matter of the patient. So I think it's very intriguing, but there's a lot we don't know yet. [00:36:40] Speaker B: Yeah, so demyelination is associated crudely, I guess, with sort of slowing of, of the propagating nerve signal. And diseases like multiple sclerosis is a demyelinating disease essentially where all sorts of things go wrong. So, yeah, that's interesting. I mean, a slower cable transmission is a very crude way of saying probably what's happening, because all sorts of things could be happening. And as you're saying, you don't really know. [00:37:14] Speaker A: Yeah. From an electrical perspective, you would expect slowing and some signal loss. Yeah. [00:37:21] Speaker B: Unfaithful transmission of signal. [00:37:23] Speaker A: Yeah, exactly. I mean, that's a very crude kind of first approximation in electrical terms, you know, what I'll say is multiple sclerosis is a much more global net for the brain. And here we're talking about what seems to be kind of very focal abnormalities. [00:37:39] Speaker B: But right at the hub, right at the superhighway hub. [00:37:42] Speaker A: Right. Not at the stimulation site, but within the network that we're treating. A few different places in that network, but there was one kind of specifically right at kind of dorsal anterior cingulate to mid singulate boundary. So one of the major white matter branches, There are a few others in the different circuits, but that's one that's been most reliable in our analysis. So certainly intriguing to us that we might be able to someday understand the subset of depression patients that are helped with this therapy and be able to describe it as a single opath, a deficit in a specific circuit that there may be all different types of therapies that we could bring to it to try and correct that deficit. Of course that is a vision for the future. But I think that's the sort of hope that gets sparked in me with the really excellent work that the imaging team was able to do there to identify these abnormalities that correlated with the activity in the brain that we were able to see and then even some of the behavior that we were able to measure in the patients because we were recording their face during their clinical interviews and their voice, and we were able to see changes in their crazy behavior that change at the same time, roughly speaking, that their brain entered stable recovery. [00:39:04] Speaker B: I was going to zoom out a little bit because there's a little bit more from the paper that I want to discuss, but let's zoom out now and talk about the sort of the current state of neuromodulation, brain stimulation with respect to just the history of where we've come from. Right. So this is like electroshock therapy back in the, I don't know what was 18, late 1800s. I don't know when all this stuff started happening, you know, and of course there were ice pick lobotomies back then, which is not neuromodulation, that is complete ablation, you know, removing parts of the brain or connections in the brain that to treat people. And those seem very crude. We look at those as if like, oh, how could you have done that at the time? And then, and then fast forward to today and you know, things are much finer grained. You're just talking about how you can kind of target areas much more specifically instead of just bathing the whole brain in some pharmacological agent without specificity to like a target area. But I still kind of think of dbs, Deep brain stimulation. Like in the Parkinsonian way. Right. You're still going in and like zapping. Well here, let me give you an example. So when I was a eye movement neurophysiologist, eye movement and cognition in non human primates, one of the ways, well, the way that we would ensure that we were in the right brain area, which in my case was frontal eye field in one was one of the brain areas that we studied. You, you would stimulate past some, some tiny currents. Well, tiny, tiny to us, but really large to the brain. Right. You would zap the brain and if they move their eyes reliably, when you zapped, you knew you're in the frontal eye field. But I always thought like, this is like a. Like what, what is that subjectively like? And what about what am I actually doing to the brain there? And it seemed kind of crude, even though it was very specific. And so where are we, do you think, in terms of the sophistication, let's say, of these kind of stimulation techniques? [00:41:11] Speaker A: Yeah, I think it's a great question and certainly I think you're bringing up an important point which is the history of developing new interventions for psychiatric disorders is a complicated one and especially around psychosurgery is a complicated one. And you bring up, you called it electroshock therapy, but electroconvulsive therapy therapy is used today and is a very effective therapy for a lot of patients. It's a life saving therapy. [00:41:40] Speaker B: But it's milder now. Right. [00:41:43] Speaker A: Certainly have been changes is my understanding. [00:41:45] Speaker B: Okay. [00:41:46] Speaker A: And also you have a better understanding of side effects and how and when to deploy it. Again, that's not my area and for. [00:41:52] Speaker B: How long and for how much current and etc. Etc. [00:41:54] Speaker A: So I don't want to speak to it too much in detail because it's not my area. I will say the broader comment that I'll make to this is, I think as we've seen the advances over the last decade, I've been very heartened to see that there is an emphasis being placed on ethics as a part of the conversation whenever science and technology is being developed, at least in the context of the things that we're doing. So I'll speak just to the Brain Initiative, which is where a lot of our work has been funded from that I was very heartened to see that, you know, ethics was a part of the discussion in the Brain Initiative very early on. [00:42:40] Speaker B: What aspects, what aspects of ethics are you talking about? [00:42:43] Speaker A: Even just bringing into the conversation that ethics has to be a part of the conversation. So, for instance, you have questions for us might be relevant to how and when are patients consented? How are you getting informed consent from someone who is in a psychiatric disorder? And so, so it's not even a specific issue. It's just the fact that's an important part of the conversation. [00:43:06] Speaker B: Yeah. [00:43:06] Speaker A: The other thing, I'll say that's been, you know, a real, a passion of mine to try and bring out, and I see it emerging more and more in the community is the lived experience voice. Oh, yeah, right. [00:43:17] Speaker B: Oh, yeah. I want to. We'll reflect on this workshop that I went to at your institution. But. Yeah, but go ahead. [00:43:24] Speaker A: Yeah, I'll say, you know, one, you know, the, the studies that we're doing right now, one of the most important things driving what we're doing is actually capturing the, the stories of the patients. You know, what is frustrating for them, the clinicians, what is frustrating for them, and then turning that into research questions. Right. [00:43:44] Speaker B: Taking anecdotes and, and using those anecdotes to. Not as evidence, which is poor science, but to turn. But to ask the right questions. [00:43:52] Speaker A: Exactly. You know, we can see that the patients are frustrated at having to try and describe in the moment whether what they're experiencing is depression or anxiety. They are saying that to us very explicitly. Right. The clinicians are reporting the challenges in resolving that. So these are lived experience stories of the people in the room at the moment. And that's, as you say, that's not data. Those are anecdotes. But you hear enough of those anecdotes and it starts to bubble up. This is an important question for us to try and address. I'll say that's something again to bring up. My partner and collaborator here, Helen Maber, that I've always admired about her is the willingness that she has just to listen to what the patients are saying and then try to understand how to pull that together and describe it and form questions out of it. So it was, despite our need to make objective measures, one of the things that drew me to entering into this partnership was the emphasis on starting from just listening to the experiences of the patients. I'll say, on another project that we're starting up where we're trying to do new behavioral experiments and measure brain body interactions and trying to understand these things that I was mentioning earlier, we have lived experience people with intracranial implants as part of the research team. Right. The chair of the council that we're starting there likes to say, you have nothing about us without us. And I think that we're hearing more and more willingness from companies and from researchers to bring the lived experience voice into these discussions. I think there's a lot more we can do with that, but I'm very encouraged that we're starting to hear those voices, the voices of our ethics colleagues and the voices of our lived experience partners be part of these conversations. So, yes, it is a complicated history, but those are the things that have me encouraged about what we're doing right now and looking toward the future. [00:45:56] Speaker B: I mean, there's a lot of talk among some groups, some circles, about needing to bring in the subjective experience of, of organisms, of humans in any question. Right. But in some sense isn't it's so daunting because there's such individual variability, variation and the. We confabulate all the time, so we have to use words to communicate and words fail us. And yeah, they never seem to fully explain what I'm trying to explain, etc. And so, so it's like you've done this great thing and you can predict these long term using this biomarker. You can predict recovery. Ain't that enough? You got to take everybody's subjective experience into account. But this is sort of the goal, right? It's like the reason for doing this. [00:46:56] Speaker A: Yeah, no, exactly. I think you're right that we're aiming toward objective measures that we can make concrete decisions about treatment on. You know, of course, someone feeling better is the ultimate goal, but we also have to acknowledge that, especially along the path to recovery, that sometimes the feelings that they're able to describe are not one to one correlates to what's happening in the biological system that needs to be managed as part of the recovery. Right. So, you know, the, the subjective stories are what get us to the questions that we have to ask. And of course, the experiences in total are the important metric if we're, you know, if we were developing a therapy that helped some objective biomarker, but at the end of it, everyone was saying, but I still can't work, I still can't enjoy my family and my relationships. I still can't live a life that feels fruitful and I feel happy with. That's not a success. Right. So there's this balance of knowing that sometimes the subjective descriptions are not operationally helpful at every intermediate waypoint, but still valuing the stories and the experiences on the whole, both to drive the questions that are the, the important things we're trying to pursue and also, you know, eventually as our outcomes. You're right. That Language is often insufficient. [00:48:19] Speaker B: Mine is. [00:48:20] Speaker A: Yeah, no, all, all of ours is. Right. And for me there are, there are two avenues that we take when our language is insufficient. One is data and trying to make things objective. And that's the path that we're on sort of here scientifically. And the other is art. Right. It's one of the great kind of redeeming things about art is it's able to express something about the human condition that goes beyond the words that we have now. That's not operationally useful in a clinical sense, but it's very important to our humanity. So for me, there are these kind of two branches when our language fails that we can use to try and describe things. And one of those has some clinical value and one I think is deeply rooted in our humanity. [00:49:07] Speaker B: Have we all right, so we know that. We've known for some time. We, not me, have known for we, as humanity, as the clinicians have known for a long time how long depression takes to treat that. It's sort of a slow wave sort of thing. Right. And you've corroborated this in terms of like the biomarker and predicting it's still slow wave. Has, has this work taught us anything new about the nature of depression itself? [00:49:37] Speaker A: I. I'm not sure yet. Right. So first of all, I want to clarify. We're working with treatment resistant depression. These are, you know, there's a very specific. [00:49:46] Speaker B: How does it differ? Can you just speak to how. Like what, what's the, what's the. Is that there's not a bright line, is there? I mean, I guess there is if you have to give it a name. [00:49:55] Speaker A: I mean, there, there are different definitions that people use, but broadly, I would say, you know, many treatments have to be tried and are no longer effective. And you know, we could, we could. [00:50:07] Speaker B: No longer effective. Like they've worked, they could have worked for a while and then it's. [00:50:10] Speaker A: Exactly. [00:50:11] Speaker B: Efficacy is lost. [00:50:12] Speaker A: So not uncommon for some treatment to work for someone for a period of time, but to start to lose its, its effectiveness. Yeah, you know, or you know, of course some treatments never work for some individuals. So while you can use different definitions, I would say they all have as a function that, that the currently approved treatments. How many medications do you have to have tried? Many of the patients have tried dozen, two dozen medications, but you have to have tried things that are not currently working for you and that has to be persistent over time. So I think in the paper you're referencing the average time in the current depressive episode was on the order of four years. Right. So it's persistent over time. It's a, you know, that's a feature of the, of the diagnosis. Right. If you look at the DSM criteria or the manual that's used to define what a diagnosis for a psychiatric disorder is for depression, it is, you know, there are two core features. One related to mood, one related to interest in activities like physical activities or. [00:51:21] Speaker B: Just activities in general. [00:51:22] Speaker A: Activities, work and activities. So what we might call anhedonia, lack of joy, lack of pleasure. There are a number of kind of secondary symptoms. You have to have some number of those. So those would be like psychomotor symptoms. So people often are slower in their movement, in their speech, things like this. So there's, you know, you have to have at least one of the core symptoms, some number of the secondary symptoms. But the point I'm trying to get at is a key part of the definition is persistent over at least two weeks most of the time. So it's a feature of the diagnostic criteria that it's a persistent condition over time. [00:51:58] Speaker B: Okay, so we're not talking bipolar, etc. [00:52:02] Speaker A: Yeah, completely different diagnosis. So I'm talking about unipolar major depressive disorder here. So, you know, by the time they're getting into a study like ours, they're treatment resistant, they've been sick for a very long time, they're not responding to anything else. The whatever is happening in the brain at that point might be and almost certainly is very different from the person who's, you know, first presenting to their doctor, saying like, I'm not feeling quite right lately, you know, and first, you know, first maybe coming on to an SSRI or some psychotherapy, some other, you know, first line treatment. It's almost certainly, you know, there's something different, you know, at that point, and I think it's yet to be understood now as we see these structural things in the brain, these abnormalities that we've pointed out, I think it's an intriguing clue, but we don't yet know whether that wind its way all the way back to something that would be identifiable in a first line depression diagnosis. I would be very suspicious of that. I think we're talking about a very specific class of patients, patients that are enrolled in our trial, so being selected through a subset of psychiatrists, maybe not even all treatment resistant depression, maybe just collections of symptoms that are being identified as being particularly approachable with this specific therapy. So I don't want to over read into what's been found, but I'm hopeful it can Help us understand something. But there's a long way to go to connect those dots. Yeah. [00:53:44] Speaker B: All right. So we all know the brain is a complex system. I had Nicole Rust on a few episodes ago, and one of the things, or sort of a major emphasis in her book is that historically we have not treated the brain in our research and, or treatments as a complex system, and that we need to start treating it like a complex system if we want to understand it and treat it better. And I've always heard and thought of complex systems as being robust. She agrees with that. But she also makes the point that they're also fragile. Fragile at, at the sort of transitions. Right, at these. Well, maybe control knobs. There are these certain control knobs that if turned the wrong way, then all of a sudden it becomes fragile, like epilepsy or, you know, things like that with runaway activity, etc. Do you think of, I mean, do you think that these. Is this is one of those sorts of control knobs that you can predict and understand and control a complex system via this? [00:54:49] Speaker A: Maybe some parts of it. And Nicole and I have, have spoken about this, and I am so grateful for the work that she has done over the last few years to be introspective about our scientific enterprise and about our foundational research and how we can think about pointing our fundamental research toward things that more effectively result in cures and therapies. Right. And. And I do resonate a lot with the dynamical systems sort of view that she's taking on it. And certainly a lot of the way that we think about what's happening specifically about a complex system. As Nicole and I have talked back and forth about this, I think it's a good question. Is a complex system controllable or is that even what you need? In this case, we're looking at specific activity in a specific circuit. Maybe not even all the activity, just some aberrant part of that activity. So it's not usually how I frame things in my mind that there's some complex system that we're trying to control. It's more like the analogy that you started to use earlier, which is, you know, this complex system is moving, it's a dynamical system, and some part of it has been stuck in some attractor basin. And, you know, with the onset of stimulation, it seems that we're kicking it out of that attractor basin. Perhaps that's something similar that's happening with short acting, you know, medications like a ketamine or something else. Short acting like a tms. But then you know, also that landscape may need to change, which you just to get to the nuts and bolts of it means changing something about the wire, changing something about the circuit. Right. When you talk about changing an attractor, you're talking about changing the topology of the network, that is, allows it to. [00:56:45] Speaker B: Have the dynamics in a different way. [00:56:47] Speaker A: You're changing the wiring somehow. And we don't know exactly what that means yet. But that's more my conceptualization that you're kicking it out of a bad place, but then you want to sustain this more typical and more healthy activity. And so the adaptation of that landscape is going to try and prevent the activity from sinking down into a negative mood state attractor like that. So I appreciate the framing. In a complex system, it doesn't tend to be how I naturally think about it. I tend to think more in terms of attractors and dynamics. But I think the questions are a good one. And it's not clear to me that we need to control the entire complex system. I'm not even sure that's, that's what I would think about aiming for. [00:57:36] Speaker B: Yeah. Okay. So the way that you're, the way that you're describing this is you kick it out of the bad state, but then you can't guarantee that it's going to land in a good state. But you have to sort of allow it to settle and then see where it lands. But there must be some protocols that increase the likelihood that it'll land in a good state or maybe that's a future endeavor. [00:58:01] Speaker A: Good question. Yeah, I don't think we know something about that, but perhaps that's the fact that these patients are going to psychotherapy therapy while they're recovering. Yeah, right, right, right. I mean, I don't know. I'm just offering that as an example of conditions that could be, you know, that could be constructed that might help that. But I don't know. I think that that's a really good. [00:58:19] Speaker B: Question because it got in that bad state somehow in the first place. And it's probably likely to just go back to that same state if nothing else changes. Right. I mean, it's gonna. That the, let's say are you're changing the, the weights of all the synaptic connections. But then it just seems like if they have, have been in one configuration for so long, why would they then all of a sudden get better and change? Right. It seems like you'd need something else to. [00:58:45] Speaker A: Well, we don't know. Right. And so we, we talked a little bit ago about these white matter Abnormalities that we're seeing. Right. So perhaps a role of that abnormality is it's allowing these basins of attraction that can pull to a negative mood state. Right. And so changes to that are something that could prevent it from being able to, you know, sink down to that type of mood state. We really don't know the mechanisms. [00:59:14] Speaker B: That's. [00:59:14] Speaker A: There's going to be a lot of work that has to be done. [00:59:16] Speaker B: So you have to be like, injecting some neuropeptides to fix the white matter or something like that. Yeah, something. [00:59:22] Speaker A: Don't know what mechanisms may be. You're putting a. You're putting a stimulator in the brain. There are probably many, many more mechanisms going on than what we've been able to measure and what we've been able to understand so far. So a lot yet to understand. But I'm, you know, it's not in my purview that I want all that activity to be controlled. Right. We still have creativity, we still have thoughts, we're still living our lives. Right. And so for me and my conceptualization is, you know, try and get a boundary up to keep it out of some bad places. People still feel sadness for very reasonable reasons. Oh, yeah, okay. [01:00:02] Speaker B: But that is a form of control. I don't think that Nicole, means, like, we need to control every millisecond of every aspect of the complex system, but to be able to move, you turn the knob and you can move it in a definitive way, that is one version of control. [01:00:17] Speaker A: Yeah. So I guess maybe your definition there is. Is it predictable what's going to happen when you apply your intervention? Right. [01:00:24] Speaker B: Yeah, that's a good way to put it. [01:00:25] Speaker A: And, yeah, I mean, we're seeing, at least in the subset of patients that are responding to this, that we see some predictability of what's going to happen. Maybe not always the timescale of the predictability and things like that, but we see some predictability. Yeah. [01:00:41] Speaker B: Okay. [01:00:41] Speaker A: I mean, that's the biomarker. Right. [01:00:42] Speaker B: It's the humble version of control. [01:00:46] Speaker A: The crudest possible version. [01:00:49] Speaker B: Can we just talk. But I want to get on and talk about your new position and some of, like, your kind of background backstory in career, because it's. You told me a story over breakfast, and I want you to fill in some holes and. And I want to. I want to. I want you to share that with the audience. But can I just ask you quickly about how you used AI in, In this endeavor in the study, because you're recording these local field potentials and then you send those signals through what's called a generative causal explainer framework, which is like a machine learning model, essentially. So maybe just high level, describe that. And then what I also just want to know is, could you have done this without AI, without modern machine learning tools? [01:01:39] Speaker A: Yeah, that's a great question. So we were recording from one specific area in a network that, that we believe is being engaged by the stimulation. So the question we started out with is, can you see anything changing in that signal that's indicative of recovery? I'll first say before we get into any talk of AI, that the most important thing we did was define what that meant. It was not an AI question. It was a question about the experience of the clinical teams that have been working on this. Right. Because let me just give you an example of something that was not what we did. You have these survey scores, right, that are being captured say every week when the patient comes in for their clinical. [01:02:30] Speaker B: I'm feeling happy, I'm feeling motive, I'm feeling motivated, I'm feeling. [01:02:33] Speaker A: So that's a score. Right. There's several instruments. The one we reported most often is called the Hamilton depression rating scale. Scale 17 question scale. So there's a number that comes out of that. Right. And so you might say, can you predict that number week to week from a brain signal? Right. First of all, that did not work well when we tried it. [01:02:58] Speaker B: Doing well, trying. By what method did you use to. [01:03:01] Speaker A: We tried several different technical techniques, you know, but you know, generalized regression, let's. [01:03:08] Speaker B: Just say, sure, okay. [01:03:12] Speaker A: But then you reflect on it for a minute and you say, first of all, okay, that's probably not even the right thing for us to be doing for the needs that we had articulated. Right. Because we've already said, listen, one of our challenges is that the movement of this survey score from week to week is influenced by all sorts of things that are not necessarily just the depressive state. Right. And this has been documented that there are, you know, there are biases, there are recency biases. It's, you know, it's a non specific measure, you know, capturing a lot of things like insomnia and. Yeah, that's right. [01:03:45] Speaker B: Which can be causal, can be a cause of and causal. [01:03:49] Speaker A: Right. [01:03:50] Speaker B: It could be circularly causal. [01:03:51] Speaker A: Yeah. Certainly a known comorbidities. Right. And also we can already collect this survey score. Right. So if that's all we need, we don't have to do any more work here. Right. If that were sufficient for our needs, just replicating that is actually not a Useful exercise. It's maybe a, a neat technical trick, but it doesn't actually get us anywhere. [01:04:12] Speaker B: Because it already exists. [01:04:13] Speaker A: They already have it. Yeah, right. And so we actually spend a lot of time in the early days of the project saying, okay, you're sort of telling us that there are things about this measure that you don't trust to the clinical teams. Right. This measure moves, you know, it's at the midpoint of the scale. It moves a little bit one direction or another. We see you not paying attention to that level of movement as you're making decisions about what to do. Maybe not paying attention is too strong, but you're not basing all of your decisions just based on that. And so we started with the question of what do you actually trust in this measure? Because you're obviously not trusting it, where a movement from a 15 to a 16 is going to influence your decision making. And after a lot of talking and mostly listening, what we came to understand is there's a lot of trust in the extremes of the measure. Measure a very high score, everyone can agree that that person is, is sick and suffering. [01:05:12] Speaker B: 15 to 17 or whatever. [01:05:15] Speaker A: Higher usually. But at the high end of this range, you know, definitely sick. We can, the clinical teams all agree about that. Once you get to the low end of that range, which on this particular scale it's below an 8, but it's not specific to. Right. You get to that end of the range, everyone can agree that, that it's. [01:05:35] Speaker B: This valley of death in the middle. [01:05:37] Speaker A: That's like in the middle, especially as people are recovering and we sort of spoke earlier that this recovery time can be very confusing as they're feeling new things. The reports can be a little unreal, not unreliable, but non specific. [01:05:52] Speaker B: They confuse anxiety with depression. [01:05:54] Speaker A: Right. Is one of the things they'll say. I can't tell. So what we came away with was, is what we can trust are the extreme points. We can all agree during time points, when this group of patients, and in this case it was the first month, this cohort of patients, a subset of them, very clear that everyone could agree that they were sick in that first month. Then in the last month, in this case, the clinical team, their combined years of experience in this, it was just a really remarkable clinical result that of that cohort of 10 patients, 90% of them were responders, which meant their scores dropped by half. That's what's the gold standard in a drug trial or anything like that. 70% of them were below the threshold for remission. [01:06:40] Speaker B: Okay, It's Amazing, by the way. I mean, that's. It's just amazing. [01:06:43] Speaker A: It's a breathtaking. [01:06:44] Speaker B: Yeah, really is. [01:06:45] Speaker A: Right. [01:06:45] Speaker B: I just want to pause there and just soak that in for a second for people to soak that in. [01:06:50] Speaker A: Yeah, I mean, that's, that's decades now. You know, that was, well, maybe at the time, a decade of, you know, when we started that cohort of experience with this therapy. And so a very experienced, very dedicated, very passionate, very smart clinical team, you know, managing, Managing this. And of course, you know, a little bit of good fortune is on your side too. Right? Like that. That may not be the sort of number that you would expect to come out of a clinical trial or something like that. [01:07:21] Speaker B: Well, it's not. [01:07:22] Speaker A: Yeah, but it's a very. It's a very fortunate situation as, as someone looking at the data, because we could get away from questions about response versus non response, and we could look at a bunch of responders and what the idiosyncratic differences were in their response. So in that cohort, there's a lot of agreement. In the last month, they're basically all, well, except for the one non responder who we can come back to. So we said that's what we can trust. So let's ignore all the data in the middle when things are very confusing and just start there. And then we're in kind of very classical, classic, you know, classification sort of problems. We have labels that we trust. We did, in this case a machine learning classifier. We tried several different kinds. It was all about the same. In this case, it was AI, you know, in the sense that it was, you know, multilayer perceptron, essentially. Not a huge one. These were not large data sets. [01:08:20] Speaker B: So this is among the AI models. This is quite. This is like the most vanilla, basic sort of model. [01:08:25] Speaker A: Exactly. Yeah, exactly. Things that we've been doing for decades before we would even have called it AI. Most of the time we would have called it just machine learning. [01:08:35] Speaker B: It's Feed Forward network. [01:08:37] Speaker A: Yeah, exactly, Feed Forward network. And we tried several different kinds, including some things that I don't know today, maybe you call AI, but we would have just called it. [01:08:46] Speaker B: I know, isn't that funny? [01:08:47] Speaker A: Logistic regression. Right. And so there's a lot of kind of robustness across these. Where I think we started to use the AI label a little more strongly was that told us that there was something changing between the two time periods when they were sick and when they were well, something in common across the group of people, because it wasn't a model trained on each individual. It was one model trained for the whole group of responders. But then there's a natural question of what is changing, right? And it's the classic sort of complaint about machine learning or AI methods, that they're black boxes and both for kind of clinical trust but also scientific understanding. There's a real pressing question here of what is changing? How do we try to understand that? And so the technique that you're referring to, Generative Causal Explainer, is a technique that we had developed previously in our lab for a type of what they call explainable AI. So it's how you take these black box models and say something about what they're keying off of. So what we did here, because a lot of the existing work was really assuming you were working with images and you know, what's the region of the image that is causing you to think that this is a bicycle versus a puppy dog or whatever, right? Not everything, but a lot of it was at the time that we did this work and we were just starting to get the early days of generative models working. So this was before the kind of big revolution in generative AI, but we were just trying to get things like autoencoders and variational autoencoders. So we're just starting to get these AI models that could generate data. So this is the basic technique. We train the black box classifier to distinguish sick from well, and then we fix that. Now that's a black box that we have. So we can put a piece of brain data into it and it can spit out a label and say, I think that brain is sick. I think that brain is, well at that point in time when that data is collected. So now with that fixed, what the Generative Causal Explainer is, is essentially a generative AI model. So in today's terms, you might think about mid journey or chatgpt, but it's trained. Instead of producing words and images, it's going to produce brain data. Okay, so it's just going to spit out brain data that statistically looks like the brain data that we had recorded. Right? [01:11:12] Speaker B: So it's like an auto encoder. [01:11:14] Speaker A: Yeah, variational auto encoder is exactly what it is. Here's the one difference. In the middle of these models, especially a variational autoencoder, there's a latent space, right? A low dimensional space. So it's a type of dimensionality reduction. You put a piece of data in through an encoder, gets down to a latent space, which is some sort of low dimensional representation, and then you're going to put it out through a decoder and get something that statistically matches back on the other side. [01:11:42] Speaker B: We have to sample from the right place in the latent space to get the right. [01:11:46] Speaker A: Right. [01:11:46] Speaker B: Well, I guess. [01:11:47] Speaker A: Well, that's what you train, right? That's part of the training. To match the statistics of the data is to construct that latent space. Okay, but now you have this latent space. So one could imagine taking a piece of brain data going through the encoder to the latent space, and now each dimension of the latent space, you could move that data point, essentially, and then you could reconstruct it and you could see what changed. You know, those, those latent dimensions don't have names, they don't have labels, but they're dialed. Right? That you could turn back and forth. So the one thing that we did differently in the training was we said, let's train this variational autoencoder to produce statistically indistinguishable brain data. But one of those dimensions in that latent space, we're going to ask it to do a special job beyond just helping to capture the statistics of the brain data. We're actually going to tie it to the output of that black box classifier. Okay, so think about it this way. This generative model is just able to spit out essentially an infinite number of fake brain data, synthetic brain data. Each one of those we're going to put through the classifier, and the classifier is going to say, that one looked sick to me. That one looked well to me. Now, that signal, we pull back into the training of the variational autoencoder of the generative model, and we ask that one special dimension to account for that decision. Okay, so now what we end up with in the training, take a piece of brain data, put it through the encoder, get into the latent space. If we then move along that one special dimension, what changes about the brain data when we reconstruct it through the decoder is something that will change that classifier's mind about whether that brain was sick or. [01:13:35] Speaker B: Well. [01:13:35] Speaker A: So put in a sick piece of brain data, you turn that one dial, and essentially the generative model is saying, how do I have to change this to make it look more like a well brain? Just like asking you chatgpt or something, hey, make this piece of text sound more professional. Right? Except we were doing this not through a text input, but right in the latent space. How do we make this brain look more well, the other dimensions, if you were to change them, would change something about the brain data, but in ways. [01:14:01] Speaker B: That didn't change the classifiers orthogonal to the wellness metric. [01:14:07] Speaker A: In technical terms, we might call those invariant dimensions because the classifier doesn't care about them at all. So now the biomarker that we actually use is that latent dimension. So we can throw that black box classifier away if we want to. We can even throw the decoder of the generative model away. We can just put the brain data through the encoder, get to that one dimension and see where we're at on that axis. That's our biomarker. [01:14:31] Speaker B: So bio. So you can. Yeah, so you have, let's say 0 to 10 or whatever. And the biomarker can be at 9, it can be at 1, it can be. And that tells you. That's the predictive component. That tells you, that tells you whether you're sick or healthy. [01:14:44] Speaker A: Yeah. Essentially asking like, what is the probability that this is a sick brain versus a well brain? Right. And that gives us the ability, because we do have the decoder, if we want it, we can turn that dial and look to see what changes. And that gives us an ability to visualize and say, as we were talking about at the beginning of this conversation, oh, it looks like this beta band signal has to increase in some relative proportion to this other signal and this other signal. So we can actually see, because it's a generative model, we can see what's changing. That would change the black box's mind about whether it was a sick brain versus a well brain at that point in time. [01:15:23] Speaker B: So this would not have been possible. [01:15:25] Speaker A: Without modern AI, I think, not in the way we were able to do it. So that both helped performance a little bit, but also gave us the ability to see what was happening. If you think without the kind of modern machine learning tools, we would have had to guess what the right feature was out of all the possible things. And remember, we're not talking about individual features. It's not just, oh, beta band change or change, as far as we know right now, it's a composite signal. It's relative changes happening across different bands. [01:15:59] Speaker B: Perhaps you'd have to exhaustively check all permutations of those features. [01:16:06] Speaker A: Exactly. And perhaps some subset of those will work out to be operational and useful all by itself. That would certainly make life a lot easier for the devices that we're using. But at least for right now, we didn't know what to look for. A data driven approach, given the important Clinical definition of what it was we were trying to look for. Let the data speak to if anything and then what is changing in the data? [01:16:34] Speaker B: So cool. That's such a cool setup. Two part question, I know the answer, but I want you to elaborate. Is this a super exciting time? But the second part of the question, is this a super exciting time for these kinds of treatments and understanding how they work, et cetera, and the potential for are we on the cusp of solving lots of these complex mental phenomenal disorders? The second part of the question is what's holding you back right now? I mean, what are the obstacles to that? [01:17:09] Speaker A: Yeah, those are great questions. I'm very excited about the time that we're in right now. I think we're seeing early glimpses of what a future can be and there are exciting things happening. I love the research that we're doing. I'm also seeing that companies like in this case Abbott Neuromodulation is starting an industry sponsored randomized controlled trial that could eventually lead to this type of therapy being an approved therapy and accessible to people outside of investigational studies. You know, our partner Medtronic continues to invest in the technology that would be important for doing these sort of things. So I think it's a really remarkable time. Even though, you know, those technologies you might think of as being crude, you know, single channels of stimulation and recording. And I think there are a lot of dreams in the neurotech industry space of going to higher resolution. Of course, I'm sure people are following the bci. You have sort of advances where many more channels of recording, you know, are being made possible and new, you know, in new techniques and technologies. [01:18:16] Speaker B: But you've shown you can do so much with solutions middle. [01:18:19] Speaker A: Yeah, exactly. So this would be a really great question for the field. We can do an enormous amount with low resolution. Higher resolution is coming online. How do we use that? Is that helpful? Right. And it's a question we get all the time. And I think until we're able to collect that data, we're not going to know. Perhaps we could do something really remarkable if we had thousands of channels of recording. Perhaps the answer is, you know, this is a global mood state and really one channel is sufficient to capture what's going on and get people, you know, get people healthy and anything other than that to overkill. We don't know yet because we've never been able to really collect that data. So I think it's a very, very exciting time. You know, from a technology perspective and from a scientific perspective. With what we've come to understand over the last decade, of course, I'd be remiss to not mention that it's also a terribly troubling time with the kind of public conversation about, about science and research and how and where that should be happening and how and where that should be supported. And so I, like many of my colleagues, are concerned that the decades of support that have led to the innovation that we're enjoying today, and that's cures for diseases, that is economic return on investment, that is the building of a workforce that can go into the nascent neurotech industry that's just exploding right now. There's certainly a lot of concern, and I think rightfully so, that that future may not exist in the way that we've benefited from over the past decades. So we'll see what reality kind of emerges over the next few years. But certainly the anxiety around that is already affecting things. I'm already seeing fewer people moving into training programs, fewer people coming to the United to be part of our training environment here, which really helps us all. So if you want to ask what's holding us back, I mean, we're human beings, we only have so many hours in the day. These types of studies are very complicated to do. They take quite a long time to even collect small amounts of data. It's very expensive and so needs both the financial and the infrastructure support to do that. But I would say the broader picture here is it's not just about one lab or one type of study. I think it's a really important time for us as a scientific community to communicate to the general public about why what we're doing has value. And there are many ways to do that. And certainly your podcast here is an enormous point of light in that need in the world. And then as a general public, we have to decide what value every different activity we could do has and how we want to put money and time and resources behind it. And I hope that science and research is something that has broad agreement, is valuable for us as a humanity to engage in. But that's a conversation that society is going to have to have. [01:21:33] Speaker B: Well, I can tell you I immensely enjoyed my time at the Interface Neuro conference that you put on at Georgia Tech and being there and then a few weeks later being at this, this Neuro AI workshop that Sean Escola put on. The sense of excitement is palpable in those arenas. And it's coming from sort of the basic research work that I do and that milieu. It was almost jarring the level of optimism and sort of contentment of people with what's going on and the possibilities, you know, so that's like really exciting. So for. From that outsider perspective, it seems fresh and new and exciting. So. Yeah, I'm excited for you. [01:22:22] Speaker A: Yeah. Got a lot of wonderful feedback about that meeting. Thanks for bringing it up. I'll say. Interface Neuro is a meeting that Rice University, their neuroengineering group, started a few years ago, and we worked on hosting a version of it here this year. They'll be hosting it next year. It's a really wonderful meeting that brings some science and technology and clinic lived experience, voices and industry all together. So have been. It's been a privilege to be part of that. One of the best comments I felt like I got was I felt optimism there in a way that I haven't in a long time. [01:22:55] Speaker B: Yeah, Yeah. I was going to make the terrible, terrible joke that me and my colleagues probably need your DBS treatment. You guys don't need it, perhaps, you know, that's not. I'm not trying to. I'm not. I'm making light of a serious condition. [01:23:07] Speaker A: But yeah, I'll say. You know, I feel that optimism. I could not be in more excited about what the future can hold for us all. And, you know, everyone comes to feelings like that for different reasons. For me, you know, just very myopically in our own life, seeing, you know, seeing a life transformed. [01:23:29] Speaker B: Yeah. [01:23:29] Speaker A: Is a really powerful thing and it really reorients, you know, a sense of purpose for you and a sense of what could be possible. Yeah. [01:23:38] Speaker B: I mean, that was one of the most. One of the most powerful things that at that workshop that you put on is just those you talked about the lived experiences of patients and you had multiple patients come on stage and share their lived experiences, which was just a treat and really makes things concrete and deserving of optimism and feeling excited, I think. [01:24:01] Speaker A: Yeah, I was very, very proud of that and worked together with partners in the neurotech space that helped make that possible. So specific, specifically BlackRock Neurotech and Medtronic really came alongside us and partnered. And if people want to hear that, some of those stories have actually just been released in a podcast from the Story Collider. It's an episode called Wired Lives. And this were people with either deep brain stimulation implants for Parkinson's for epilepsy, or people with brain computer interfaces. So some of the BCI pioneers that were, you know, in the early research studies or continue to be in research studies, and they came on stage, told their personal stories. And now I think just about a week ago was released as, as an episode called Wired Lives from the Story Collider podcast. [01:24:55] Speaker B: I'll link to that in the show notes. Also, people can just search it up. Also, we don't have a ton of time here, but you told me a story at breakfast, excuse me. About a month ago, at one of these workshops. And I want to. So we don't have hours and hours here, which I'm sure you could spend like talking about this, but you come from a non affluent background in Michigan, like we said, apparently playing hockey and. [01:25:26] Speaker A: But didn't play, but was a fan. [01:25:27] Speaker B: Okay, Big fan. All right. Big fan. And these days you just became the inaugural director for the new what's called the Institute for Neuroscience, Neurotechnology and Society at Georgia Tech. Congratulations. [01:25:42] Speaker A: Thank you. [01:25:44] Speaker B: So I don't know where we should start here, but I kind of want to. Well, I also want to know like, how your job is going to differ now than, you know, just running, just running a lab, you know, and doing all the research that you did, how the nature of your job will differ. But one of the things that you communicated to me was how much you, how important it is to you to, to help people around you and to form collaborations and to lift other researchers up and to communicate what you're doing to society. And you gave me the reasons for why you feel that way. I was hoping that you would kind of share those reasons with me and with the audience. [01:26:32] Speaker A: Yeah, happy to. So, as you sort of mentioned, I grew up in a fairly rural area in Michigan, a place where, you know, it's. It's not always common or expected that, that people are going to go on to college or go on to scientific careers. I'm not sure I knew scientist was a career one could have. Growing up first generation to go to college. In my family, you don't hear that much anymore. Yeah, yeah, but that was true. You know, grew up in a place where I saw, you know, I saw people that were in difficult circumstances. My grandmother's member of the local Native American tribe there certainly saw, you saw difficult circumstances to break out of and was fortunate, you know, that there were people around me that saw, you know, a talented kid that they wanted to help get opportunities but didn't really know what those opportunities should. Should be. So I think part of my story to convey is that I had no idea what I was doing. I stumbled. I took, I want to say wrong terms because I think that are all valuable, but I Did not take a straight line to what I'm doing. I went to college. I checked the box that said computer engineering because people told me I was going to be an engineer. I had played with a computer a few times and thought it was kind of cool. And so those words together, no idea what it meant, but I'll check that box. [01:27:55] Speaker B: But did you, at that time, did you have ambition? Were you driven or were you. Were people just telling you how smart you were and you thought, well, this is what I'm supposed to do? That's kind of what happened to me. I didn't have an ambition. People were just. I sort of did it because that was the story people told me about myself. [01:28:09] Speaker A: Yeah. I don't know that I had a specific vision or ambition. You know, there was a lot of that. Right. You know, people telling you that you're good at something and you should do it. You know, I think that there's a certain amount of, when you see that you're good at something, you lean into that because. Helps you feel good. [01:28:27] Speaker B: Yeah. [01:28:27] Speaker A: Right. Grew up in, you know, kind of a farming family and, and blue collar jobs where, you know, hard work is just how you survive. Right. And so I kind of grew up in a way that, you know, just your, your default mode is you put your head down and just work as hard as you can at something. And so ambitious in that sense that I felt like, you know, I, I want to go do something. [01:28:54] Speaker B: That's how you get it done. [01:28:55] Speaker A: Yeah. But I didn't. It wasn't a specific vision. I would have told you that I wanted to go design computers or something, but I didn't. So I went to college again, just kind of picking a place. I ended up at the University of Michigan, which was an amazing choice, but it wasn't a very informed choice. You know, I went because I had a few friends that I happened to know that were there. They had a marching band that looked like it would be a lot of fun to be in. I did a lot of arts and performing work, ended up auditioning for the music school there my second year. So I ended up doing a dual degree in music and engineering and ended up being there six years. Right. But that was fortuitous because I was still figuring out who I was. You know, it was both. It was a time of newness where I was meeting people from all over the world. I had been, you know, just in this one, you know, kind of rural region and so didn't you didn't travel much? No, not at all. Yeah, not at all. I had only ever been a few hours away from where I grew up. Okay. And so it was, it was eye opening to me to be around people that were different than me. At the same time, there was a level of familiarity and comfort that I hadn't quite experienced before because it was, it was a lot of people that were, you know, they were thinking about thinking for a living and these kind of, you know, your careers in science and engineering and things that I just hadn't been exposed to. And so it was this disorienting time where it felt both very foreign and comfortable and familiar in a way that I hadn't experienced experience. And so that, that time was valuable for me to figure out something about who I was. As a practical matter, let me start doing research and start teaching for the first time. Because those last couple of years just from a credit hours perspective, slowed down quite a bit and so it created space for some new experiences. [01:30:47] Speaker B: You're teaching as an undergrad? [01:30:49] Speaker A: Yeah, they. I had, after four years, I had essentially finished my engineering degree. I was just working on the music degree degree. And so the engineering school essentially let me be a ta. [01:30:58] Speaker B: Oh, cool. [01:31:00] Speaker A: For, for one of their low level classes. And it was a transformative experience for me. So, you know, I went through this time thinking a research career, you know, might be what I wanted to do. I didn't really know much about research, but I knew it was essentially a prerequisite to be a professor at a good place. [01:31:20] Speaker B: You didn't know, you didn't know that it intended thinking for. I mean, you mentioned the thinking part, but like that, that, that would be, I mean, it's part of a draw, right, is you get to be creative and solve problems, solve hard problems cognitively. [01:31:34] Speaker A: Yeah, I'll say I feel that way now. At the time I would have said that is a necessary component to be teaching at a major university. And that's really what I want to be doing. [01:31:47] Speaker B: The teaching part or the being a professional professor part, being called being a professor part. [01:31:53] Speaker A: But mostly I thought of the teaching. When I thought right there was something very gratifying and that's, you know, tied into. I was a camp counselor for many years and so I was seeing the impact you could have investing in young people's lives. Right. And so these things were coming together. The ability to invest in people, the kind of technical aspects that were very cool to be a part of the educational aspects. And so I did not yet have the spark for, you know, the creativity through the research really driving Me, but I was saying, you know, I think the questions are most interesting are probably ones related to the brain. I don't think I can, you know, at the time it was really like sensory perception was really driving me. Right. You can imagine through the music degree. And I said, I don't think I can understand that by studying music. I think I need to understand the brain. But I'm obviously not going to go be a neuroscience major. I haven't had a biology class since the ninth grade grade. I'm not on that sort of track. I'm kind of vocationally an engineer. So I wonder if I could use these tools of engineering to study the brain. And it was, you know, this is a long time ago. Right. So that was not a common sort of track. But I was able to find a few people, including the person I ended up studying with, Don Johnson, that was a card carrying engineer. Don was at Rice and had had really been an electrophysiologist while an electrical engineer and then had essentially become one of the early generation of computational neuroscience scientists. And so ended up going to work with him. Being surrounded by all this incredible engineering talent. Could take the neuroscience classes from the medical center across the street at Rice and get this really rich interdisciplinary mix of the neuroscience, the engineering tools coming from this arts background that I think just kind of broadened my perspective. And so it wasn't intentional, but I think I stumbled through this path. Path that has combined this milieu together that has ultimately been beneficial for me. Right. To be able to think about what creativity means in the context of research and how to give and take feedback and all the things that you learn as an artist and mixing it together with, with the science and the, and the technology. [01:34:00] Speaker B: Is that why you value interdisciplinary approaches so much? [01:34:05] Speaker A: 100% it is. You know, my, my major parts of my personality have been forged in ensemble work. Yeah, yeah, right. And so I've never quite fit in this mold of like a single PI, you know, with their single R01 doing their own work. That's always felt like an uncomfortable stance to me. And you know, certainly a lot of people warned me that that's not a way to build an effective career. [01:34:30] Speaker B: Why? Because you're going to be one of a hundred authors on every paper or. [01:34:33] Speaker A: Yeah. Just how do you, you know, if you aren't going to kind of own a space and develop depth in that and be known as the expert in that space, it can be difficult to succeed. And I don't want to minimize the real challenges with that, but I think that was never a comfortable posture for me to take. I always felt more comfortable and think that my skills really are best suited to trying to bring different areas together and translate languages across areas. It's just something I've developed a fluency for. So if I was going to do this job, I think the only way that it felt good to me was to try and sit in these interdisciplinary spaces. And if I couldn't be successful with it, I couldn't be successful do the. [01:35:19] Speaker B: Job of running a lab. [01:35:21] Speaker A: Yeah. I don't think I knew another way to try and do it that felt right. Natural to me. Right. So that was kind of the journey that I started out on in theoretical neuroscience and really thinking about sensory systems through building some kind of technological tools and then wanting to have more applied impact for a number of reasons, including seeing the effect of neurologic disorders in my own family, brought me to wanting to think about the clinic and through an interdisciplinary meeting of clinical colleagues up the road at Emory with the engineering colleagues here. We were trying to drive those interactions and. And found myself in conversations that felt really resonant and felt like we had something to bring and to offer, especially in the space of psychiatry. Right. So I think that's kind of a whirlwind tour through it, but it is why I kind of feel so passionately about the thing that we're trying to do here is so complicated that. [01:36:25] Speaker B: It'S. [01:36:25] Speaker A: Bigger than any one of us. It's bigger than any one thing that we can hold in our own expertise. And so bringing together, together, if scientists of different stripes, engineers of different stripes, clinicians of different stripes, the lived experience, voices, like the magic for me happens where all that comes together. And so I know it's not the typical path for a professor, but I've come to love the research part of it in a way that it was not the spark that started me. And even when I started as a professor, I would still say, you know, I enjoyed the research, but it was. It was part of Emilio here. And I think it's. While I still enjoy teaching, I love teaching. I think that I've grown into a space where the creativity and the impact that we've been able to see through the research and especially through the interdisciplinary part of it is what's really animating to me. [01:37:20] Speaker B: We kind of skipped over your grad school days along your path here, and you had mentioned to me that there were some challenges along the way there as, I mean, I don't know, a graduate student or someone who obtained their PhD who does not have war stories from their graduate studies. But I mean, or questioned, you know, whether they wanted to continue existing, whether they wanted to continue the program, et cetera. So, but, but you mentioned you had some that, that made you question your choices. What, what, what was going on there. [01:37:50] Speaker A: Yeah, absolutely. I'm glad you brought that up because I, I, I skipped over it in the story and that was probably an oversight because I think it's a really important story to do tell. You know, we see these very polished bio sketches and CVs of people, and it looks like it was a linear path. It's often not. [01:38:06] Speaker B: Right. [01:38:06] Speaker A: I've always admired, you know, the folks that have kind of the shadow CV up where they write every grant that they applied for, that they got a rejection and that. So you're right. So grad school was a particularly difficult time for me, maybe one of the darkest kind of periods of my life, because it was, it was a time where maybe the first time that I wasn't succeeding at what I was doing. It was a time where I had kind of a very specific notion of what was coming next. Right. I had this dream of becoming a professor and was kind of day by day seeing that slip away. [01:38:38] Speaker B: Because you weren't, because you felt like you were failing or I was not. [01:38:42] Speaker A: Felt like I was, like I was not thriving. [01:38:44] Speaker B: Yeah. [01:38:46] Speaker A: And you know, there are a number of sort of reasons for that. You know, one, one, technically, I was not the strongest in the program. That's just kind of objective fact. I have other skills that aren't measured on tests, but I technically was not particularly strong. It was also so new to me and the environment was so foreign that I just wasn't making progress. My specific issue was we had these very abstracted ideas, but I was having trouble turning them into concrete research questions and just felt like stuck. And my advisor, who is wonderful and I have a great relationship with and I, I love him, hugely influential to me. He didn't know how to get me unstuck from that. And so I was just sitting and I would, I would come home to my wife, I was newly married at the time, and I would say, I didn't do anything today. And she would say very graciously, like, you just don't see the value in what you're doing yet. That would become apparent with time. You just have to stick with it. [01:39:45] Speaker B: I mean, you didn't accomplish anything today. [01:39:48] Speaker A: And I would correct her and say, you didn't hear me right. I didn't do anything today. I Stared at the wall of my office for seven hours because I didn't know what to do. [01:39:58] Speaker B: Crushing, right? Yeah. [01:39:59] Speaker A: And so, yeah, I almost left the program. I remember going and sitting, you know, in the office of my kind of closest friend there and saying, I'm not sure how much longer I can do this. Right. And so it was very dark. I ended up just, you know, I think it's an important part of the story to tell. I ended up in a cardiologist's office at something like 24 years old with him saying, what are you doing here? Right. You know, with essentially health trouble that I had induced myself through the stress that I was feeling. And it was clear it was damaging my young marriage at the time. It was damaging me. Right. And so there had to be a major reset. And, you know, there were a number of things that helped. One was I brought more mentors into my circle, both at Rice and beyond. You know, people like Rich Baranek and Bruno Olshausen, a theoretical neuroscientist that I would eventually postdoc with before I came here. And I had to be sort of forceful, not because they weren't willing, but everyone is just very busy. And I essentially sat down in Rich's office one day and wouldn't leave until he wrote to Bruno, who I didn't know at the time, and invited him out to visit because I didn't know what else to do. And through widening that circle, I think, you know, it let me see the parts that I wasn't learning. I had to see this process of formulating a research question one. One time, right? I had to see it executed one time up close. And then something clicked, right? And then I could do it. Somehow, I wasn't getting that from where I was sitting. And by expanding my mentorship circle, I was. And so I know a lot of people's, you know, if they have trouble in grad school. So I had a terrible advisor or the environment wasn't great. I was blessed. It was an amazing environment, amazing mentors. I was just stuck in some attractor that I couldn't get out of. And I think I had a wonderful support system of friends and family. My wife, Cara, is an angel sent here to Earth to protect me from myself. And she had a really shrewd piece of advice, which is, would you sit down and write a mission statement for yourself? Not what job do you want to have, but what are you about? And I did that, and it was actually a very freeing exercise because it let me kind of step away and say, okay, I have this vision of what my career is going to be. That may or may not happen, but I've articulated what I'm about, and no matter how that career works out, I can still be about those things. I might have to find a different career path. Right. To express those things. And maybe it's a collection of different ways, you know, through volunteer or different careers, that I can do those things. [01:42:39] Speaker B: But that was a central identity that you could come back to as a. [01:42:43] Speaker A: It was. And you could make decisions from. And it was very comfortable to be able to separate my success on a particular career track with my own identity, to basically separate my identity from my professional success. And then I could be much more rational about it. I could say, listen, I grew up in a certain way. I know what bottom looks like. I'm educated, I'm employable. My family's not going to go hungry. I can take care of my basic needs, which wasn't always a game given for me. So I know what I'm about. I can lean into these things in many different career paths. This one career path is still my aspiration, but I know that I can still be about who I am. And that was a, you know, one of the greatest gifts that my wife has given me, in addition to just the grace of sitting with me in that time. That was so difficult. It was. It was freeing in a way that just kind of unlocked. Unlocked something for me. So I was glad to be able to navigate out of that time. I was glad to be able to see that my skill set is actually better suited for this job than it is for the job of being a grad student. [01:43:55] Speaker B: Yeah, that's odd. You got through the. [01:43:57] Speaker A: Yeah, yeah, yeah, I know, right? And I know that that's not most. [01:44:00] Speaker B: It's usually opposite. [01:44:01] Speaker A: I know, I know. But I think that is true for me. I'm better in this role than I was as a graduate student. And it let me kind of see that. That that was going to be the case. It let me, again, kind of divorce my identity from what I'm doing, which has been a helpful skill in this job, because these are jobs that can really eat you up. They will take every amount of time and energy that you can give them. And so being able to draw boundaries around the job has been an important part of trying to stay healthy in it. And you're using the, you know, a rooting of my identity in something that's not my professional success, or at least I'm imperfect at it. But that's my aspiration and that's my goal and that been part of the toolbox that I've tried to use to keep healthy through a career of this job, which can be quite hard. [01:44:54] Speaker B: All right, Chris, I know we're up against the limit here. I know that you have kind of a hard out, but thank you for sharing that personal story and your personal stories along your journey. Thanks for sharing also the research. Congratulations on the new position. We did not get to talk about Neuromatch, which I will link to in the show notes and people can learn more about but hopefully we can talk about that another time and hopefully we can share another breakfast sometime at another workshop somewhere. So anyway, thank you for coming on. It's been a pleasure talking. [01:45:24] Speaker A: Yeah, thank you Paul. It was a privilege to be here. Thank you again for all the work you're doing trying to bring a light in the world to what's happening in the science and research world. It's a great thing that you're doing, so it's a privilege to be here. Thank you. [01:45:44] Speaker B: Brain Inspired is powered by the Transmitter, an online publication that aims to deliver useful information, insights and tools to build bridges across neuroscience and advanced research. Visit thetransmitter.org to explore the latest neuroscience news and perspectives written by journalists and scientists. If you value Brain Inspired, support it through Patreon to access full length episodes, join our Discord community and even influence who I invite to the podcast. Go to BrainInspired Co to learn more. The music you hear is a little slow jazzy blues performed by my friend Kyle Doniphan. Thank you for your support. See you next time. [01:46:32] Speaker A: It.