IntegraPose: A unified framework for simultaneous pose estimation and behavior classification

Computational ethology often relies on fragmented pipelines that separate pose estimation from behavior classification. We present IntegraPose, an open-source, GUI-driven toolkit that unifies these tasks in a single YOLOv11x-Pose model. Benchmarked against a foundational DeepLabCut model, IntegraPose achieved comparable keypoint accuracy, reduced prediction jitter (improved temporal stability), and substantially faster processing speeds (∼72-73 FPS vs ∼32 FPS). This high-throughput, synchronized analysis enabled the discovery of a context-dependent motor deficit in neuropathy-prone C57BLKS/J (BLKS) mice: a phenotype evident during the biomechanically demanding wall-rearing behavior but not during simple locomotion. The intuitive interface streamlines the workflow from annotation to advanced analytics. By producing a single, time-locked stream of pose and behavior, IntegraPose facilitates hypothesis-driven studies of behavioral mechanisms.