ASiDentify (ASiD): a machine learning model to predict new autism spectrum disorder risk genes.

The article focuses on the development and application of ASiDentify (ASiD), a machine learning model designed to predict candidate autism spectrum disorder (ASD) risk genes using 18 genomic, RNA, and protein features. Trained on 1,152 known ASD susceptibility genes, ASiD identified over 1,300 candidate risk genes, including more than 300 not previously predicted, with six key features—loss-of-function intolerance (LOEUF), synapse protein localization, and gene expression in astrocytes, excitatory and inhibitory neurons, and non-brain tissues—being most informative. Modified versions of ASiD applied to RNA-binding proteins (RBPs) and chromatin regulators revealed distinct feature patterns, suggesting different biological pathways linking these functional groups to ASD. The model's predictions showed significant overlap with other ASD gene prediction models and were enriched for genes downregulated in ASD cases, supporting the excitatory/inhibitory imbalance theory of ASD, while also highlighting the potential importance of gene paralogy in ASD risk.