Cognition and behavior, Novel methods and technology in neuroscience

Publication date: 22/05/2024

Authors: Goodwin, N.L., Choong, J.J., Hwang, S. et al.

Journal: Nature Neuroscience volume 27, pages1411–1424

Commentary: In this study, Goodwin et al. present SimBA (Simple Behavioral Analysis) that is an open-source platform used to classify complex animal behaviors using pose estimation data and supervised learning, all through an accessible, GUI-based interface. Above all, SimBA is designed to bring explainable machine learning into behavioral neuroscience. Indeed, a central innovation of SimBA is the integration of SHAP (SHapley Additive exPlanations) methodology, which provides detailed insights into how specific features such as movement, posture, or proximity contribute to behavioral predictions. This level of transparency is critical for understanding model decisions, comparing different classifiers across labs, and refining behavioral definitions. The authors validate SimBA through a series of experiments. In chronic social defeat stress (CSDS) assays, they identify sex-specific differences in aggression-related behaviors, with females showing longer and more frequent bouts of attack and pursuit. In resident-intruder (RI) tests, they observe a shift from exploratory to aggressive behaviors over time, with SHAP revealing how environmental context alters feature importance. Additionally, comparisons between rat and mouse aggression reveal species-specific behavioral signatures. In summary, SimBA’s ability to quantify and visualize feature contributions makes it a powerful tool for reproducible and interpretable behavioral analysis and classification. By turning behavior into a measurable and shareable resource, SimBA supports even more rigorous and collaborative neuroscience.

Commented by: Filippo La Greca (09/06/2025)

DOI: https://doi.org/10.1038/s41593-024-01649-9