This Seed is a new collaboration that combines PI Allen-Blanchette’s expertise in neural network modeling with Co-PI Datta’s expertise in studies of bacterial active matter. Their goal is to develop a new class of graph neural network model to predict and control collective behaviors in bacterial active matter. Specifically, their approach will leverage techniques from neural relational inference, to infer latent communication and belief graphs from observed quantities such as the physical characteristics and trajectories of individual bacteria via a novel message passing framework. The inferred communication and belief graphs will then be used to determine the community dynamics that are governed by a nonlinear opinion dynamics model selected for its interpretability and ability to capture the behavior of a broad class of biological decision-making systems. Combined, these strategies will allow for the discovery of bacterial interaction graphs, ultimately allowing for forecasting of active turbulence, directed motion, and spatial structuring in bacterial collectives. Additionally, because bacteria are excellent models of more complex systems of multiple actively-interacting agents (e.g., mammalian tissues, colloidal microswimmers), this work will also yield a principled approach to alter and rationally control collective behavior in active materials more broadly.

Principal Investigators:

Christine Allen-Blanchette (MAE)

Sujit Datta (CBE)

Seed start and end date: January 1, 2024 - December 31, 2025