GEESE: Genotype-aware End-to-End Spatio-temporal Embedding for Behavioral Phenotyping
Researchers have developed GEESE, a novel deep learning framework designed to automate behavioral phenotyping in genetic animal models. This end-to-end system learns directly from 3D pose data, eliminating the need for manual feature engineering and improving reproducibility. GEESE has demonstrated superior performance in classifying behaviors and predicting genotypes across multiple autism-associated genetic models, identifying genotype-specific movement signatures. Additionally, the project includes HONK, an interactive tool that allows researchers to perform phenotyping using natural language commands. AI
IMPACT Automates complex behavioral analysis, potentially accelerating genetic research and drug discovery.