Focus on research: Dr Claire Gillan, Adapt
Dr Claire Gillan is associate professor of psychology at Adapt, the Science Foundation Ireland research centre for AI-driven digital content technology. In this interview she talks about using Big Data and bigger populations to make sense of mental health.
Tell us a little bit about your academic background.
I did my undergrad in psychology in UCD. At first I wanted to be a clinical psychologist then accidentally landed in research. I got in touch with someone at Cambridge to do a little bit of research and he said ‘why not stay for a PhD? It’s only three years.’ I didn’t realise I could do that. Within the first few months I realised this was the thing for me rather than being a practicing clinician.
I did my PhD working on Obsessive Compulsive Disorder trying to figure out what goes wrong in the brain that puts people at risk. We did a lot of work on the basic brain systems that support how we learn habits or automatic behaviours. People with OCD rely on this system more than they should and it causes them to pick up compulsive behaviours that are really hard to shake.
I had a great time doing that work and that led on then to a Fellowship at New York University where I worked for three years and started to get more interested in data analytics and computational methods. Then I got my position here at Trinity first as assistant professor and now associate professor.
Where did the interest in Big Data come from?
It came about from a research question. When I was doing my PhD we were using a ‘status quo’ kind of approach to understanding mental health – getting a bunch of maybe 20-30 people with a diagnosis and comparing them to some healthy controls.
The problem with this kind of research was that the sample sizes were much too small to get an understanding of the brain systems we were studying. We’d start to see what was happening in the case of OCD then we’d start to see the same sorts of effects in other patients, for example, with social anxiety or patients with schizophrenia. That caused me to really question the work I had been doing. It shone a light on the implications of studying these very small samples where patients tend to have combinations of many different symptoms and we can’t statistically disentangle those when you only have 30 people in your study.
At NYU I wanted to figure out if what I had been studying was putting people at risk for compulsive behaviours and the only way to do it was to have a sample of over 1,000 people with varying levels of symptoms so we could statistically separate them. That got me using online methods to try and gather a large sample in a relatively short space of time. So we did a study where we got 1,400 people and most of them were recruited in 10 days, as opposed to back in my PhD where it took two years to get 30 people.
That really kickstarted things for me in the technology space both in the lab and in the more sophisticated analytics. Ever since we’ve been doing research on a much larger scale and using these sorts of methods to understand what we now know to be a very complex landscape of mental health problems.
In using Big Data you would be reliant on mathematical models where you make an adjustment and look for the effects downstream.
I think of it more as if we have the research questions first then think about what kinds of tools we can use to answer them most comprehensively. Sometimes you don’t need Big Data at all but for the things we’ve been working on the last few years about what is the specific brain mechanism linked to a specific diagnostic feature in any mental helth problem like being compulsive seems to be. Can we really identify the brain processes that leave people vulnerable to that. That’s where Big Data methods have been useful. As we try to appreciate that complexity we find we need them more to answer the kind of questions we do in my lab.
Machine learning is a good example. We use it because a focus of our work now is trying to see if we can use these basic insights to deliver something of clinical value, or can they tell us about the future rather than the present. We do a lot of work trying to characterise different kinds of people but the most powerful thing we do is ask ‘can we understand if someone presents to a clinic what’s going to happen to them in a few years. Are they going to get better under a certain treatment or are they going to relapse after a certain treatment?’
For those problems we found machine learning to be really useful.
Your signature project Neureka brings together app development and citizen science.
We’re four years into the project, two years since launch, we’ve about 25,000 registered users. We’re really happy with how that’s going and have a really deep and rich characterisations of those citizen scientists. They tell us a lot about themselves directly, and indirectly by playing the games on our app. We’ve had to learn a lot of things that are outside of our wheelhouse. One thing we’ve learned is that grant funding doesn’t seem to cover things like user experience or making the app pleasant for people to use. We try to strike that balance now between applying for funding to do science verus the day-to-day stuff of all the things you need to support this project. It’s a different animal to a lot of the other research we do in the lab and we’ve just had to learn a lot about things that go beyond the science.
Tell us a little about how the app works
Anyone can download it. You get some information up front about what the data is used for, how it’s processed, what the limitations of the data we gather are, then they are presented with what we call ‘science challenges’. The idea is that at your own pace and level of time commitment or interest you can complete as much or as little as you like but everything you do helps propel science forward to understand the causes of disorders of the mind.
Some of our initial funding was for work looking at how to keep the brain healthy with age but we’ve since gotten other funding for basic science about how people learn habits and also work on problems around predicting clinical outcomes.
We use the data for several different projects, some fuelled by citizen science on the app and other times we’ll use the app in a specific randomised study where we’ll get people into the lab for assessments.
Finally, we’re starting to let other groups use it for their own research projects as well. We’re setting up collaborations in the UK, the US where all kinds of research can be facilitated through this one app.
When it comes to using citizen science were you concerned about relying on people to be accurate and honest?
It’s a concern with all kinds of data collection. A big thing we’ve been concerned about for a long time is all the research participants in psychology studies are female twenty-year-old undergrad psychology students. That’s an enormous problem and some of the other issues pale in comparison. When we go online to get data we do see more distraction, that’s a bit of an issue. However when we pay people to come in and lie them down in a scanner they fall asleep. I don’t think you get more honest answers in person. It might work the other way around as people don’t want to reveal stuff face-to-face.
People are often incentivised by the financial gain whereas online people’s motivations are aligned with the research. The reason they’re there is to advance a particular area and because it’s on their own time they’re much less likely to be actively dishonest.
We look for convergence across different methodologies. We’ll do a study online then try and replicate it in that setting, but also with different populations and under different conditions. In one of the most recent papers we published we compared the data in our app from citizen scientists to a group of students that we paid to do the exact same thing and we got the same results across the two data sets.
SFI labs named after their founders tend to have a sense of community and purpose. Do you see yourself as a mentor?
One-thousand per-cent. That’s how it works and one of the reasons why the labs are often named after the principal investigator is because the topics change over time but not the person leading that lab. If that person goes away that lab does not continue.
Mentorship is extremely important in academia in general and in particular our lab. We take the approach that everyone is coming in to learn new skills and it’s our job at the end to make sure they’re going into whatever industry they want.
Subscribers 0
Fans 0
Followers 0
Followers