ICM
Unravelling behavioral variability: a data-driven approach
Animal behavior varies widely, both within the same individual over time and between individuals. While often overlooked, this variation reflects hidden control variables and mechanisms that were shaped by evolution. For example, variation in behavioral traits can help populations withstand environmental change, while atypical motor patterns in neurological disorders may offer clues for personalized therapies. Comparing such complex behaviors is difficult. When dynamics are nonlinear and unfold over multiple timescales, standard metrics based on summary statistics often miss meaningful differences. To address this, we introduce a framework that encodes multiscale dynamics to compare behavior from data. By modeling nonlinear dynamics probabilistically (using transfer operators inferred from time-series data), we define a distance metric that captures behavioral differences across timescales. Tailored to finite, noisy datasets, our approach identifies principal axes of variation and enables rigorous clustering of individual trajectories. We demonstrate this framework in various biological systems, including bacterial chemotaxis and larval zebrafish locomotion, where the inferred axes of behavioral variation reflect underlying physiological variables and developmental histories.