Data-driven discovery of long timescale behavioral strategies during sensory evoked locomotion

Survival requires the animal brain to generate flexible and adaptive behavioral mechanisms on multiple timescales. To identify the internal states that modulate behavior across scales, we require an understanding of how short timescale behaviors are chained to generate long sequences. We leverage the larval zebrafish to characterize such sequences because of the availability of high-resolution, high throughput recordings in various sensory conditions.



Larval zebrafish naturally swim in short timescale burst like motions called bouts. We construct a maximally predictive state space by stacking consecutive bouts, and then study the time evolution of state-space densities through transfer operators. Their spectral decomposition reveals slowly decaying modes corresponding to stereotyped long timescale behaviors.



We find two long-lived strategies in larval zebrafish locomotion lasting tens to hundreds of seconds which occur naturally during exploratory behavior in light – “Roaming”, which causes fast changes in orientation and “Cruising”, which is dominated by forward locomotion. Our analysis reveals how stimuli modulate such long timescale behaviors by either triggering or driving the fish into roaming or cruising. We discover a clear structure in larval zebrafish behavior at long timescales and how it is modulated by external stimuli, enabling discovery of the internal states regulating behavior.