Spectral mode dynamics of C. elegans

The roundworm C. elegans exhibits diverse behavioral states and motor programs. Recent advancements in imaging platforms have enabled the collection of large amounts of data to record videos of C. elegans motion over long time periods. Constructing models for animal motor programs and connecting these models to behavioral states and neural patterns is a major current area of research. In this work, we present a general method for learning motor programs in a spectral representation. The method is robust to experimental noise and imposes necessary physical constraints and symmetries of the system. The learned models are sparse and physically interpretable. We utilize this learning approach in concert with clustering techniques to identify distinct motor patterns in C. elegans.