"I find beauty in the everyday things in science."
An accomplished neuroscientist with an interest in the neural basis of adaptive motor behaviors, Mackenzie Mathis recently joined EPFL’s School of Life Sciences as the Bertarelli Chair in Integrative Neuroscience at the Brain Mind Institute.
Tell us a bit about your background.
Before journeying to EPFL, I was a Rowland Fellow at Harvard University. That fellowship allowed me to start my own lab right after my PhD in 2017, and then we moved the lab here this summer.
What do you work on?
We're a systems neuroscience lab. We're very much interested in how neural circuits control behavior; in particular, we're very interested in how animals can adapt to quick changes in their environment. If we think about the massive amount of neurocircuitry that's involved – everything from the cortex to control dexterous movements of the hand, all the way down to the spinal cord – there are a lot of interconnected “loops” in the brain. Trying to understand how this is smoothly and robustly orchestrated is a long-standing and challenging question that we try to tackle with modern techniques, using mice as our model system.
Our lab is also quite heavily invested in new machine-learning tools. We've been building tools to track the mouse's movements with high fidelity. It’s the equivalent of doing motion capture, but without markers—i.e., to make movies like “Avatar” people typically wear MoCap suits. But mice don't like wearing costumes, and it turns out many other animals don't either, such as cheetahs or aquatic animals. So, there's a lot of utility in being able to do this non-invasively without markers on animals.
In 2018, we were the first to combine human marker-less deep learning motion-capture to do this in a way that required very little manual data annotation for animals. Namely, you define which key points you want to track on the animal, and then leverage these deep neural networks to do this automatically for you.
That's been a huge part of our research in the last few years, not only for tracking tiny mouse hands, but it’s been expanded it into a lot of different fields, which is quite exciting for us as developers to be a part of. Our software package for this, DeepLabCut, has been downloaded so far 160,000 times in the past two years— making it quite widely used tools in neuroscience, ethology, and even in clinical trials; recently, the Atlantic and Bloomberg wrote about it as well.
What are some great moments in your research?
Science isn't full of milestones where you get rewarded constantly. You have to inherently love your work. The moments that are most exciting for me, personally, are probably the ones where you're in a room, you're doing an experiment, and you're the first person in the world maybe to see or discover something. Of course, now as a Professor this extends to seeing this enthusiasm and joy in your students.
I find beauty in the everyday things in science. I think without that, it would be very hard to be a scientist.
Our lab’s a philosophy is that we “pop the champagne” when we post a pre-print/submit a paper, not necessarily when it's accepted (but, let’s be honest, we perhaps do so at both points). Nonetheless, that point at which you decided you have something of value that you want to put out into the world, and you're excited for your peer-review feedback, is the most rewarding one. There should be a lot of pride and enjoyment in that moment of finishing a work packet.
Why did you choose EPFL?
I was very excited about the research community here. EPFL is truly one of the top places in the world, where you have amazing computer science, engineering, and neuroscience – especially strong motor neuroscience, which is the sub-field we work in. Being embedded in such a community that has such incredible research groups really made EPFL the top choice.
What are your future plans?
We're interested in how animals can dynamically adapt their behavior. And that adaptation takes many forms. We've embarked on a research program that looks at this from both the learning of “skilled-games” in mice perspective, to more ecological behaviors, where animals interact with their daily environment. The overall theme is that we want to understand such intelligent systems from several perspectives: how do neural circuits enable this adaptation? Can we imbue some of these adaptive rules into actual artificial (AI) systems? Thus, the integrative approach in our lab is between natural intelligence and artificial intelligence. We're trying to learn from both of these fields to make better AI.
The other key perspective, and an area we want to work towards, is therapeutic applications. I believe we can't completely engineer our way to new medications or to rehabilitation strategies for the motor system without understanding how all these different, and quite complicated hierarchical-control loops, are interacting. Therefore, we take a “basic-science” approach, but we hope that building tools like DeepLabCut and new computer vision tools that can be used in therapeutic settings, is a step in that direction. Moreover, we hope that by working towards understanding the neural circuits, we can try to use that knowledge in a more targeted method for future therapeutics, which of course might be years off.
But I think that's another advantage of being at EPFL, which fosters a thriving entrepreneurship and cross-discipline culture. We really hope to interact with labs that are also working in these directions to build scalable and robust solutions.