Dr. Carsen Stringer
Group Leader HHMI Janelia Research Campus
Postdoctoral Fellow HHMI Janelia Research Campus
PhD in Computational Neuroscience Gatsby Computational Neuroscience Unit, University College London
As a sophomore engineering student at the University of Pittsburgh, Carsen Stringer was captivated by her nonlinear dynamics course. In particular, she found chaotic systems theory fascinating, especially when applied to the brain. She was moved by the idea that small variations in the timing of a single neuron’s activity could change the activity of all the neurons in the downstream circuitry. Inspired to learn more, she changed her major to math and physics and began working on computational research projects with her professor, Dr. Jonathan Rubin. With his mentorship, she learned how to simulate passive dynamic walking, a system where two legs walk down a shallow slope, using only gravity as a power source.
When taking neuroscience courses in her senior year, she set her next goal: to pursue a PhD in computational neuroscience. She chose the Gatsby Computational Neuroscience Unit at University College London. The program appealed to her because she was deeply interested in the cutting-edge neural simulations they were running, as well as the program’s emphasis on machine learning, which she predicted would be useful for her career. It also helped that she had already become acquainted with the U.K. during a study-abroad semester at Oxford.
In her first computational neuroscience project at Gatsby, Carsen used simulations to generate chaos in networks of neurons. Around then, she met Marius Pachitariu, then a fellow Gatsby student and now Carsen’s partner. Marius was analyzing auditory data from rodents using recurrent neural networks, and he suggested that she try simulating the data. In fitting network simulation models to neural data, Carsen became aware of the limitations of the process. She explains that when you use network simulations to model a system, there are many assumptions you must make. Most importantly, there are a lot of free parameters in neural networks, all of which can be adjusted in your model--inhibitory weights, timescales, and intrinsic currents, to name a few. She found that the data often do not sufficiently constrain these parameters. In other words, there can be too many ways to arrive at an answer. For instance, in the paper they published on this work, they concluded that inhibition was the strongest driver of firing rate changes across brain states and sensory modalities. Although inhibition was a powerful factor, Carsen notes that changing other parameters could also maintain a working model. “There’s either not enough data or not enough constraints,” This realization led her to think more about learning how to perform her own experiments.
Around then, Marius had begun large-scale neural recordings in rodents with Matteo Carandini and Kenneth Harris’ group, and when looking at the data, he and Carsen assumed they were observing a dynamical system with highly recurrent activity. That is--the activity within the circuit was driving the circuit. However, they found multiple concurrent patterns of activity in the brain, which did not harmonize well with their recurrent network models. Instead, they found that some external input drives this neural activity, and that those external inputs correspond to behavioral states. Inspired by this finding, Carsen dove into experimental neuroscience. Co-advised by Matteo and Kenneth, she used two-photon calcium imaging to investigate how behavioral states are represented in brain-wide activity. She found that the behavioral states that map onto brain activity are actually multidimensional. That is, instead of the neural activity reflecting a single dimension of behavior, like whether the mouse is whisking or not, Carsen actually found that the mouse’s precise facial expression, a multidimensional measure, was represented in neural activity!
Soon, it was time for Carsen and Marius to decide on their next career step. The duo landed positions at Janelia Research Campus--Marius as a group leader, and Carsen as a postdoctoral fellow, where she was co-advised by Marius and Karel Svoboda. She made the decision to stay in the field of visual neuroscience because she felt that was where she could “make the most progress,” a testament to her philosophy and dedication to the group effort of scientific research. In a similar vein, she is committed to open science and open source projects. She has worked extensively on several open source projects: Cellpose, suite2p, rastermap, and facemap. These projects are all documented on Github and thus, the scientific community contributes to improving the code. A major benefit of working at Janelia is that Carsen can use time she would otherwise spend on grant-writing or teaching to help develop tools that others rely on. She is grateful for the privilege to make her contributions to science by way of tool development and open science.
After two years as a postdoctoral fellow, Carsen landed a group leader position at Janelia. At the time of our interview, (January 2021), her lab was six months old! Despite the pandemic, she has had a blast mentoring students remotely. She is grateful for her past mentors and hopes to apply what she learned to her new trainees. Just as Jonathan Rubin allowed her to explore her own ideas, she strives to mentor people in a way that preserves their freedom of thought. Another key mentorship style she appreciated was the excitement and passion from Matteo Carandini and Kenneth Harris. In their lab, research was fun and exciting--this made for a very encouraging environment, which resonates with Carsen’s style of approaching her questions.
Her lab is focused on two major ideas: understanding how visual computations work, and understanding how behaviors are represented in the brain. She explains that classical and deep network models are not the whole picture when it comes to understanding visual computation. As for the behavioral representations, she is still very curious about the global signals driving activity patterns, a thread she has been weaving since her PhD. For Carsen, the iterative nature of modeling and experimentation is crucial to a healthy scientific process. This is why she publishes all her datasets and encourages everyone to do the same. She prefers her work to be data-driven as opposed to purely hypothesis-driven. In this way, she can see what the data tell her, and make models based on the data. Hypotheses are, of course, always important. But we should be careful not to frame everything under one model, or one hypothetical framework. And, she cautions against being too dogmatic about what we think we know; it can always change. She appreciates the open-mindedness of experimental labs and spaces, and certainly embodies this healthy attitude in her own approaches and philosophy of science.
Find out more about Carsen and her lab’s research here.
Listen to Daniela’s full interview with Carsen below or wherever you get your podcasts!
