Abstract
Autism spectrum disorder (ASD) is a highly heterogeneous neurodevelopmental condition caused by many genetic mutations, each conferring modest individual risk. Despite its genetic complexity and clinical burden, no effective treatments targeting core symptoms are available. Although ASD is diagnosed as a single disorder, accumulating evidence suggests that it comprises biologically distinct subtypes, yet how genetic risk maps onto phenotypic substructure remains poorly defined. Here, we apply zebrafish behavioral phenomics to systematically link ASD risk genes to functional behavioral subtypes in vivo. Using an unsupervised deep learning-based behavioral profiling platform, ZeChat, we characterized the behavioral repertoires of zebrafish mutants for 11 high-confidence ASD risk genes. ZeChat analysis revealed distinct behavioral phenotypes for each mutant that was classified into three behavioral subtypes (Clusters 1-3) via hierarchical clustering of their behavioral fingerprints. By integrating mutant phenotypes with a large neuroactive drug phenomics database, we identified subtype-specific correlations with serotonergic, dopaminergic, and opioid receptor-modulating compounds. Using an orthogonal social preference assay, Fishbook, we further demonstrate subtype-specific pharmacological rescue of social deficits, including selective rescue of the Cluster 3 mutants arid1b, adnp, grin2b, pten, and the Cluster 2 mutants pogz and shank3, by μ-opioid receptor antagonists. Together, these results demonstrate a scalable in vivo framework for behavioral phenomics-driven classification of ASD risk genes into functional subtypes and discovery of targeted rescue strategies.