DeepFaceMouse enables scalable prediction of large-scale brain activity from facial dynamics in mice.

DeepFaceMouse is a high-precision, scalable deep learning framework for analyzing orofacial behavior in mice and predicting large-scale brain activity from facial dynamics. Built on a pose-estimation pipeline with optimized training, inference, and data-handling workflows, it enables robust tracking across diverse behavioral conditions and neurological disease models. Systematic benchmarking demonstrates improved tracking precision compared with existing approaches, along with substantial gains in computational efficiency, particularly on GPU-enabled systems. Importantly, these improvements translate into more accurate prediction of distributed cortical activity across multiple brain regions. By linking behavioral measurement quality to downstream neural signal reconstruction, DeepFaceMouse provides a reproducible and scalable framework for studying brain-behavior relationships. This approach establishes a functionally grounded benchmark for evaluating behavioral modeling methods and supports applications in both basic and disease-focused neuroscience research.