Autism research in genomics is hindered by fragmented data resources that are hard to combine in statistical models: While SFARI’s SPARK and Simons Simplex Collection cohorts provide extensive whole-exome and whole-genome sequencing (WES/WGS) data, these are not integrated into widely used genomic portals. Additionally, SFARI-funded single-cell RNA sequencing (scRNA-seq) datasets, which offer insights into cell-type-specific gene expression, are dispersed across publications and archives, lacking centralized access. This fragmentation impedes both experimental and computational analyses and it is unclear how much combined analyses can increase the accuracy when predicting autism gene and variants impact.

To address these challenges, this project aims to (1) integrate SPARK/SSC WES/WGS variants into the UCSC Genome Browser, harmonized with sample metadata and linked to SFARI Base, facilitating accessibility for researchers; (2) centralize SFARI-funded scRNA-seq datasets in the UCSC Cell Browser, incorporating quality control, metadata annotations and integration into a unified reference atlas using methods like Harmony to enable cross-study analyses; (3) extend our PM1 multimodal AI model to integrate SPARK variants and single-cell expression data, generating variant and gene-level importance scores, cohort embeddings, and cross-modal imputations, with outputs surfaced through UCSC portals while ensuring privacy safeguards.

Leveraging established infrastructure and expertise, this project will deliver accessible, integrated, and AI-enhanced autism genomic resources. By bridging genetic variation, cell-type-specific expression, and autism phenotypes, it aims to identify key genes, variants, and cell types underlying autism risk, providing a new framework for integrating different genomic data types in autism research.