Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with a strong genetic basis.The role of de novo mutations in ASD has been well established, but the set of genes implicated to date is still far from complete. The current study employs a machine learning-based approach to predict ASD risk genes using features from spatiotemporal gene expression patterns in human brain, gene-level constraint metrics, and other gene variation features. The genes identified through our prediction model were enriched for independent sets of ASD risk genes, and tended to be differentially expressed in ASD brains, especially in the frontal and parietal cortex. The highest-ranked genes not only included those with strong prior evidence for involvement in ASD (for example, TCF20 and FBOX11), but also indicated potentially novel candidates, such as DOCK3, MYCBP2 and CAND1, which are all involved in neuronal development. Through extensive validations, we also showed that our method outperformed state-of-theart scoring systems for ranking ASD candidate genes. Gene ontology enrichment analysis of our predicted risk genes revealed biological processes clearly relevant to ASD, including neuronal signaling, neurogenesis, and chromatin remodeling, but also highlighted other potential mechanisms that might underlie ASD, such as regulation of RNA alternative splicing and ubiquitination pathway related to protein degradation. Our study demonstrates that human brain spatiotemporal gene expression patterns and gene-level constraint metrics can help predict ASD risk genes. Our gene ranking system provides a useful resource for prioritizing ASD candidate genes.One approach is based on the concept of guilt-by-association, i.e., assuming that genes that confer risk for ASD are likely to be functionally related, and that they thus converge on molecular networks and biological pathways implicated in disease (12, 13). For example, one study showed that ASD genes with de novo mutations converged on pathways related to chromatin remodeling and synaptic function (14). To leverage these functional relationships, several studies have explored integrating known risk genes using a protein-protein interaction (PPI) network to identify novel genes involved in ASD (15-18). However, a PPI network is built upon general protein-protein interactions without reference to tissue or cell type specificity, and this approach may not fully capture the brain-centric functional relationships among ASD genes. Accordingly, a brain-specific network-based approach, which considered relationships within the context of the brain, was proposed to predict ASD genes, (19, 20). Studies employing this paradigm, however, did not consider the dynamic patterns of gene relationships during brain development, thereby limiting their potential for discovery, given the possibility that genes might only be functionally related within a specific developmental stage. Evidence for this comes from Willsey et al., who showed, using 4 spatiotemporal gene expression data from human ...