Scientist Stories: Marta Jankowska, Nucleotides to Neighborhoods

All right, hi everybody.

My name is Martha.

I am a researcher at UC San Diego.

I am a health geographer, and if you're wondering what a geographer is doing here, I'm gonna honestly tell you I have no idea.

I don't know how I got here, to be totally honest.

It's like one thing led to another, and before I knew it, I was like researching genomics, and I was like, what's happening? So I'm kind of gonna take you on this journey that I've gone through, And I want to tell you about this study that I'm leading called Nucleotides to Neighborhoods, which fit in really nicely with one of the first slides we had, which literally was a nucleotide all the way up to a neighborhood. And I was like, I need that. So you're gonna have to send me that, 'cause that would be the perfect image for this. So I've always been really interested in integrating different types of data, and always with a geographical bent, since that is my background.

And I've always been interested in health, and obesity and cancer risk have been kind of one of my main focuses.

So I'm not going to get into all this.

We know all this. It's very complex.

I didn't have the complexity slide. I usually just throw up like a ball of yarn because to me that's what it is.

But, you know, there's all these different factors that are going to that we have to consider when we look at obesity.

So my mentor and I, when I first started my postdoc, I worked on this study called Community of Mind, and we just wrapped up data collection for this in 2017, and it's 602 people that we actually purposely selected from a diversity of environments in San Diego County. That's always— A big problem that we find in a lot of studies is that you usually select people based on outcome, and then it turns out that all those people with that outcome live in the same type of environment, and then how are you supposed to study environmental variability if there is no variability? So we purposely took people from a variety of obesogenic and non-obesogenic census tracts in San Diego, and then we just measured everything. I'm sorry, we totally broke challenge 2, I think that was the one, right, where it's like get all the data. We did that. So we— sorry.

So we did clinical visits, and from that we got blood and urine and a couple other things, biometric measures.

We put sensors on them. That's a big interest area of mine is GPS and accelerometer sensors, also heart rate and other types of sensors. But for this, study, it was GPS and wrist accelerometers, so we got both movement during the day, but then also sleep patterns at night from the wrist accelerometry. We did so many surveys, like, these people hated us by the end of it, but so, you know, there's all the demographic, there's all the perceptions, depression, clinical history, just the gamut of questions. Then we did two 24-hour food recalls to try to get at some of the nutrition variability, And all of this is done in the name of trying to come up with better measures of environmental exposure interacting with behavior to understand cancer risk that is being caused by insulin resistance and obesity. And the cool thing about this study too is we purposely tried to get at least a half sample of Hispanic Latinos, so we have pretty good ethnic diversity in this sample. So, that's my slide, which I feel is a little lamer than yours, but I'm very interested in environmental data. And so, as we— as these people wear the GPS units for about 2 weeks of time, I get really great information about where they're going. And then, if you are interested in incorporating environmental characteristics into your studies, please come talk to me because I have pretty much everything that there is to be had that is freely available in the United States. And then if you want to do it in San Diego, then I just really have a lot of data. But so we're looking at, you know, all kinds of sociocultural-based characteristics, built environment, hazards, assets. I've started getting— my postdoc is doing a lot of stuff with social media data, which can be geotagged in a lot of cases. And so we're getting, like, really cool data sets that look at like perceptions of vaccinations and all kinds of stuff. So there's a lot of really amazing datasets out there that are pretty—

they're not easy, but they are free.

And so as long as you have the manpower, they're able, you can get it. So what did we do with all this data? We decided we wanted more. I'm so sorry. So I put this picture up, and I'm probably breaking copyright rules here, but because what we're really getting into now It's not necessarily big data. We don't have a lot of people, but it's very deep data.

And so this is what I constantly feel like.

I feel like I need to hold a string as I lower down into the layers and layers and layers of this data. So we started this subset in a group of this PQ study called Nucleotides Neighborhoods. And for these people, basically, we added more data onto—

you're not gonna report me, are you?

Okay. We added more data to the Community of Mind dataset. So, we— out of the 602 participants that we recruited, we had 526 that consented to doing genetic testing on their blood samples.

And so, out of those, we started—

I'm so sorry.

So we are in the process of doing all of these different epigenetic, epigenomic-based measures.

Oops.

We are working with Rob Knight's group to do—

or we already did. We collected microbiome in a subset of 202 of them, and then I'm trying to get funding to add ATAC-seqon to it, but we'll see if that actually happens. So why did we do this? I'm very fascinated in that ball of yarn and trying to really, when we talk about a system, and I really do, this is I think a great point of discussion to really say like where's the hypothesis and where's the data?

But at the same time, I feel like particularly in the realm of behaviors and where they fit into all of this, we have kind of really not great understanding.

And so I really wanted to get this kind of comprehensive, this really A to Z information about people. And so I'm gonna show you some of the successes we've had and some of the challenges that we've had with this data. So this is a starter table. Whenever someone comes to us and says they wanna work with us, which by the way, shameless plug, we are really looking for collaborators. Obviously there's a lot of data here and obviously I have a PhD in geography. I don't know like 90% of what to do with this stuff.

So we need help with this data.

But when someone first comes to us, we kind of give them this table because this is kind of getting at the first characteristics that we're gonna be pulling out of the data that we have.

And so as you can see, you can just start drawing lines between all of this, right?

You can just start drawing lines. You can start drawing spirals.

You can start drawing circles.

There's a lot going on here that you can start trying to connect. It's a small sample size, but it's meant to be exploratory.

Exploratory.

And so we've done some interaction network modeling. We had one of our students—

actually, this is more of a review of her just trying to look at literature that's connecting all these different pieces.

And then we're starting to put it all together in interaction network modeling.

But the questions are, you know, how do these parameters relate to one another?

And then how do they interact? And then I— that question of what do we do with the unknown interactions? Like, I feel like there's just so much here still that we don't know yet. And earlier when someone was talking about that you first draw the model and then you kind of draw in the question marks, that really resonated with me because there's so many question marks for me with this data.

One of the things we've done is start exploring with the microbiome data.

Just looking at dichotomies, looking at do they spend a lot of time outdoors, do they not spend a lot of time outdoors? Do they eat a lot of fresh fruits and vegetables, do they not?

And so we're starting to see some pretty interesting differences.

Like for example, people who spend more time outdoors have more diverse microbiomes. And we're kind of starting to see a lot of these like really fascinating connections. But what's really interesting with this is we haven't even begun to look at this minute-by-minute level data, which there could be some really interesting thresholds here that we haven't even begun to scratch the surface of. And I had a conversation earlier that kind of speaks to the challenge of this, because really, if we really wanted to kind of see these changes, you know, we'd probably have to be getting microbiome data like weekly or something like that to start to try to match the temporality of this behavioral data to temporal changes in these other types of data that we're getting. So kind of along that line, where I'm really excited is looking at chromatin.

I'm learning what chromatin is.

I still don't have like a really solid grasp of it, but what I do understand is that chromatin is a— the way that it binds with our genome can lead to rapid changes, right? So people can— or not people, chromatin can bind, and then if something happens, it can quickly unbind. And so you can get pretty immediate reactions.

Maybe not immediate, immediate, but it's a pretty rapid change.

So the reason I put this up with this data here is because in traditional— not traditional, but in a lot of epidemiological studies that look at environment, what we're really talking about is the home, right? And what we're talking about is long-term exposures.

We're saying that someone's lived here for years, they've had some kind of exposure for years, and it's consistent because it's around the home.

Well, now that we're moving into these GPS-based measures, we can really start to get these assessments of in real time and in quite a lot of detail what they're truly being exposed to throughout the day. And then how do we connect that back to what's going on within the body? How does that get reflected in the body? I think chromatin might be a really exciting way to do that because we can get these kind of more rapid changes in an environment. So particularly air pollution, I think would be a really good example of trying to get at that.

As you can see, we haven't had many successes yet because we're still kind of trying to wrap our brains around how to deal with all this information.

So the challenges, we've had a lot of challenges. We do have a great multidisciplinary team, and as we've talked about a lot already, this is essential to doing any of this kind of research. The challenge I've really found though, it's not—

so siloing's an issue. It's not that people don't want to work with you. It really though is like getting people to start thinking outside of their lane.

And that really has been difficult because, like, when we work with the Microbiome Center, we work with the Genomics Center, you know, all they want to do is run the analysis that they know how to do. They just want to run their data as they always run their data. And when you kind of are like, well, I don't know what to do with that, or I need it in a different format, or— that's when it really starts getting tricky.

And then finding the right people to do that is really hard.

Is a challenge. And more often than not, it's gonna be either a PhD or a postdoc, 'cause the younger kind of generation, or the younger people are the ones who are a little more into trying new things.

So training, that's the next problem. I, like, the training issue is a huge one.

And I know a lot of fields have moved into, like, kind of these institutions and more problem-based models rather than, you know, you're gonna go train in microbiology and you're gonna go train in computer science. Like, do we need to start thinking about creating actual academic institutions like departments that are this particular thing that have one of every single thing that you might need to think about a systems approach?

Data integration methods.

There's progress within all of these fields, but there's just the scale of the data, the different scales. That's what gets me, and that's what I was talking earlier with David about, is that we have incredibly deep data in our genomics about one person at one point in time, and then we're trying to join that with really, quote unquote, long data, as in temporally, weeks of someone's data, minute by minute at a time. Which then maybe you're going to want to join to life course data with different points of their life. I mean, so it just—

my brain hurts. And it—

but the— like we're going to need to figure out how to do this.

And I don't think collecting— okay, contrary to what you might think, I don't think collecting more data may—

is like the immediate answer. I think we're going to have to get creative and doing things like synthetic datasets And maybe assigning them attributes in various ways might be one way of doing that. We talked a little bit about workflow development, but I think that's really important. And I think it's so, so essential.

So for someone like me to get into your field, I mean, the things I've had to learn—

no geographer should ever have to learn.

But my point is simply—

It's not easy, right?

Nothing is easy here.

There's no, like, I'm gonna load up a software and click a button. I'm gonna put my data into that software and click a button. So when we start talking about sharing 20,000 lines of code, no, I don't want anything to do with that. I would like something that has some kind of workflow.

Has anyone ever used Kepler?

It was developed by a woman named Ilkay Altintas at San Diego Supercomputing Center.

And it's not easy, but it starts taking these developed pieces of code and putting them into kind of like—

if you don't have kids and you play the kind of code building thing where you have blocks of code, they're like chunks, and you just start dragging them onto a screen. And as you put them together, they start making a process. That's what Kepler can do. So thinking along the lines of something like that, where people who are outside of this field can say, "Oh, okay, this code for processing the genome," which I know is not that easy, but, you know, "This code for integrating the microbiome," like, where you can start kind of drag and dropping.

That's going to take so much work, but that's kind of my dream for trying to get more people who are out of this field to be able to use be into it. So I'm—

okay, I'm almost done.

Okay, so my last thing is recruitment and sample size.

This is a really big issue, which again, I was really excited to learn about the synthetic data, because I think that's a really big option here. But, you know, our study of 600 people, getting people to enroll to give up this much data was so challenging, so challenging. So how do we even get people to agree to this? And then also, how do we make sure we're being representative of different ethnicities as we try to collect this kind of information?

How do we generate adequate sample size estimates?

A lot of grants that I've been applying to, I get reamed on this constantly. It's like, where's your power coming from? How do you calculate the sample size?

I don't know because no one's ever done it.

So that really to me is like a, I don't know how to get around that.

But it brings me back to one comment I was gonna make about Dr.

Hu's comment earlier about generating, about coming up with ways to do either RCTs or understanding how we actually figure out causality. Well, in my mind, sample size is very closely tied to that because you might be talking about a sample size of 1. How do you determine causality in a sample size of 1 when you get into this kind of deep, deep data, right? You're not gonna get population-level estimates for this kind of data. You're not gonna get 300,000 people to sign up for this kind of study.

So we need, like, what is the scale that you're doing your study in?

And how then do you, what are the methods that we need to come up with to actually say something meaningful about what we've measured?

How to deal with deep data computationally. Yeah, okay. That's it. Okay.