Northwest Labs B103 and Zoom
Title: Bridging scales in C. elegans behavior: from posture to pirouette to path
From molecular motions within a cell to patterns of neural activity to organism-scale behavior, the living world is a teeming, dynamical dance of interconnected components. In the behavior of animals, advances in machine vision now provide access to this dance with unprecedented quantitative detail. But how do we translate rich, posture time series into new understanding? Here, we combine delay embedding with Markov modeling to extract important behavioral dynamics from partial, time-series measurements. We introduce maximally predictive states, a symbolic encoding designed to maximize short-time predictive information. Transitions between these states provide an accurate reconstruction of ensemble dynamics, which we use to reveal timescale separation and long-lived collective modes. Applied to the behavior of the nematode C. elegans, we show that our ensemble dynamics effectively bridges sub-second posture fluctuations and long-range effective diffusion in centroid trajectories. Further, we find that “run and pirouette” behavior emerges directly as a single, long-lived mode, and we identify new behavioral states.