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

Area of Science:

• Genetics
• Neuroscience
• Computational Biology

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with a significant genetic basis.
• Identifying all ASD risk genes is challenging due to genetic heterogeneity and limitations of current sequencing methods.

Purpose of the Study:

• To develop a machine learning model, ASiDentify (ASiD), for predicting novel autism spectrum disorder (ASD) risk genes.
• To identify key genomic, RNA-, and protein-based features associated with ASD risk.

Main Methods:

• Developed ASiDentify (ASiD), a machine learning model utilizing human genomic, RNA-, and protein-based features.
• Applied ASiD to predict candidate ASD risk genes, identifying over 1,300 potential genes.
• Utilized logistic regression models to analyze features specific to RNA-binding proteins (RBPs) and chromatin regulators.

Main Results:

• ASiD successfully predicted over 1,300 candidate ASD risk genes, with over 300 being novel predictions.
• Key predictive features included mutational constraint, synapse localization, and gene expression in neurons and astrocytes.
• RNA-binding proteins (RBPs) and chromatin regulators were identified as particularly implicated functional groups.

Conclusions:

• The ASiD model effectively predicts novel ASD risk genes, expanding the known genetic landscape of the disorder.
• Distinct biological pathways may link different molecularly functioning risk genes to ASD.
• Further investigation into RBPs and chromatin regulators could reveal specific mechanisms underlying ASD pathogenesis.