This project combines deep learning models of gene regulation with the whole-genome sequences (WGS) and phenotypes of the SPARK cohort to test the hypothesis that de novo structural variants (SVs) disrupt neurodevelopmental chromatin interactions and gene expression in autism spectrum disorder (ASD). Rare and de novo variants in more than a hundred protein-coding genes have been linked to ASD, with genes in unique molecular pathways characterizing different phenotypic classes of ASD. Noncoding variants have been less deeply characterized in ASD, despite mounting evidence that they play a role in neurodevelopmental disorders. In particular, there has been very little investigation of noncoding SVs even though their effect sizes are expected to be larger than those of single nucleotide changes. Identifying casual SVs in the noncoding genome would reveal new genetic mechanisms and could identify additional ASD genes that lack truncating or protein-damaging variants but increase risk when differentially expressed during neurodevelopment.

To address this gap, the Pollard lab will use their state-of-the-art models that predict how noncoding SVs change chromatin-based gene regulation in the developing brain. De novo SVs detected in WGS of ASD probands but absent from their unaffected parents will be scored with SuPreMo, the lab’s software pipeline for ranking thousands of variants based on deep learning predictions of how much each variant disrupts cell type specific chromatin interactions and gene expression. To test the hypothesis that de novo noncoding SVs contribute to ASD risk, SuPreMo scores will be compared between probands and unaffected siblings. High-scoring proband SVs will be linked to genes and tested for associations with phenotypic classes of ASD, molecular pathways, and cell types. These analyses will quantify the contribution of noncoding SVs across the phenotypic spectrum of ASD and identify new risk loci in which gene regulation is a likely causal mechanism.