Studying naturalistic animal behavior remains challenging. While machine learning advances enable limb localization, extracting behaviors requires understanding spatiotemporal pose patterns. B-SOID, an open-source, unsupervised algorithm, identifies behaviors without user bias. By training a classifier on clustered pose pattern statistics, it improves processing speed and generalizes across subjects/labs. A frameshift alignment paradigm overcomes temporal resolution limitations, providing sub-action categories and kinematic measures from a single camera. This is crucial for studying rodent and other models of pain, OCD, and movement disorders.

Alexander I. Hsu, Eric A. Yttri