Author: Sam Sooter | Majors: Physics, Mathematics | Semester: Spring 2024
I’m a physics and mathematics major, and I received a SURF grant for Spring/Fall 2024
to fund my research in Dr. Woodrow Shew’s lab in the physics department. I plan to go
to physics graduate school, then pursue a postdoc, and eventually start my own lab as a
physics professor, focusing on collective behavior in biological systems with many
interacting parts (like the brain).
I started working in the Shew Lab just over two years ago after taking Dr. Shew’s
Thermal Physics class. The Shew Lab studies the idea that the brain operates close to
the edge of a phase transition, a state called “criticality.” Close to criticality, large,
scale-free fluctuations emerge. The main reason for caring about criticality in the
context of the brain is that these scale-free fluctuations are thought to confer a number
of functional benefits. The brain executes computations over a broad range of
timescales, from responding to sensory input (on the order of 10 ms) to deciding what
word to type next in a blog post (on the order of 1 s), and criticality is a mechanism for
generating such a diversity of timescales.
My research in the Shew Lab has taken two complementary paths. First, I
developed a new tool for estimating proximity to criticality from single-cell resolution
recordings of local populations of neurons. This tool is based on the renormalization
group (RG) from statistical physics. The idea of RG is to transform a model (i.e., a rule
for generating data) into a new model at a coarser scale, and to iterate this procedure to
get a series of effective models at coarser and coarser scales. If this flow in the space
of models terminates at a fixed point, then we say the model is at criticality, although
the fixed point must be “non-trivial” in a technical way I won’t describe here. This can in
fact be taken as a definition of criticality. For a simple class of linear, Gaussian models
called autoregressive (AR) models, we can analytically identify all the fixed points (all
the different kinds of criticality) and conditions on the model to flow into them. Starting
with a recording of neural activity, we can fit it to an AR model and then, using the
results of our RG analysis, calculate the distances to the basins of attraction of all the
fixed points. This approach is a huge improvement on existing techniques for
estimating proximity to criticality, as it is both more direct and, crucially, more time
resolved. This work emerged from a collaboration between the Shew Lab and math
professors Cheng Ly (VCU) and Andrea Barreiro (SMU) and will be on BioRxiv and
submitted for publication a few days from now.
My second research path, strongly intertwined with the first, is linking the
animal’s behavioral state (i.e., whether the animal, usually a mouse, is running, sleeping,
stressed, or what have you) to its brain’s proximity to criticality. The tRG (the “t” here
stands for temporal) tool that I developed gives us access to proximity to criticality at
the fast timescales on which behavioral state fluctuates, allowing us to make this link. I
found that mouse cortex is closest to criticality at an intermediate level of alertness,
when the mouse is neither too active nor too drowsy. Consistent with this, Antonio
Fontenele, a postdoc in the Shew Lab, used the tRG tool to show that cortex of awake,
head-fixed mice running on a wheel is closest to criticality at an intermediate running
speed: at the behavioral extremes of running too much and not at all, cortex deviates
from criticality. Future work will delve further into the connection between criticality and
behavior.
I presented a poster on my research at the NEXTEN conference earlier this month
at WashU’s Center for Computational and Theoretical Neuroscience. The premise of the
conference was to discuss what the “next ten” years of neuroscience might look like.
Most of the speakers suggested the interplay between AI and neuroscience will be most
important. Maybe so, but I think criticality as an organizing principle for brain function
will also star in the next ten years.