MEAM Seminar: “Engineering Inputs, Decoding Brain States – Microscale Mechanics and Machine Learning for Whole-Brain Neuroscience”

About this event

The nematode C. elegans has a fully mapped nervous system of 302 neurons, and yet predicting how its brain turns sensory input into behavior and how experience affects the process remains challenging. Much of the difficulty is an engineering problem. The ability to control what the animal senses with precision, and to read out the activity of the whole brain while the animal moves and deforms would greatly enhance such studies. In this talk I will show how we use low-Reynolds-number flow and soft-material mechanics to design these experiments. One example is a microfluidic clamp that uses a pressure drop across a tapered constriction to hold the head still for imaging while the body behaves freely and chemical stimuli switch on and off. Another is a pneumatic membrane device that delivers calibrated touch under different conditions and across the lifespan. I will also describe self-supervised machine learning that tracks and identifies neurons in deforming 3D recordings without manual annotation. Finally, I will show what these tools reveal in an learning context where training reorganizes neural dynamics across the whole brain, pushing low-dimensional brain states toward those that drive the learned behavior, while leaving basic sensing intact.

From Mechanical Engineering and Applied Mechanics