A general framework for discovering long-timescale dynamics in animal behavior

Animal behavior spans a wide range of timescales, from repetitive movement patterns like locomotion and grooming to long-term behavioral states influenced by the internal state of the animal, such as circadian rhythms and hunger. Accurately quantifying changes in these internal states is key to understanding the upstream neural and biochemical processes. Thus, it is necessary to separate long-timescale behavioral dynamics from the faster dynamics occurring at shorter timescales. While recent advances have allowed us to isolate these long timescales from animals like worms and fish with relatively simple body plans and a single movement timescale, this approach generalizes poorly in practice when applied to legged animals like flies or rodents. We extend this approach by demonstrating that a transfer operator, approximated from signals in the time-frequency domain, effectively separates the slow dynamics across a variety of animals from worms to flies to rodents. Our results identify known long timescale states in worms and flies and provide a general framework for how to extract these types of dynamics from postural trajectories.