Scaling Up Social Behavior Studies: Real-Time, Large-Scale and Prolonged Social Behavior Analysis with PrecisionTrack

Large-scale ethological behavioral studies can provide novel insights into the neuronal processes underlying complex and social behaviors, potentially opening new mental health research avenues. However, conducting such studies remains a tedious and time- consuming process, often requiring manual annotation of predefined ethological features. Although researchers are aware that current protocols constrain the potential scope of their experiments, they have limited alternatives. Despite recent advancements, automatically monitoring complex behaviors in socially interacting animals remains difficult. Even gold- standard tools struggle to reliably track more than a few interacting animals, forcing researchers to conduct short, small-scale social experiments in confined environments. Here, we present PrecisionTrack, designed to remain accurate and reliable, even when tracking large groups of interacting animals over prolonged periods. PrecisionTrack outperforms both DeepLabCut and SLEAP in crowded social settings, maintains its tracking performance over time, and supports real-time, online responsiveness. Researchers can therefore use it to gather ethological data on the fly during ongoing experiments. We also introduce the Multi-animal Action Recognition Transformer (MART), which can be seamlessly integrated with PrecisionTrack to extract behavioral insights beyond spatiotemporal localization, including individual actions and social interactions. The entire system is open source and accessible through high-level APIs. Users can follow step-by-step guides to configure, train, test, and deploy their own social behavior monitoring systems.