Inferring genetic variant causal networks by leveraging pleiotropy

Genetic variants have been associated with multiple traits through genome-wide association studies (GWASs), but pinpointing causal variants and their mechanisms remains challenging. Molecular phenotypes, such as eQTLs, are routinely used to interpret GWAS results. However, much concern has recently been raised about their weak overlap. Taking the opposite approach with PRISM (Pleiotropic Relationships to Infer the SNP Model), we leverage pleiotropy to pinpoint direct effects and build variant causal networks. PRISM clusters variant-trait effects into trait-mediated, confounder-mediated, and direct effects, and builds individual variant causal networks by cross-referencing results from all traits. In simulations, PRISM demonstrated high precision in identifying direct effects and reconstructing causal networks. Applying PRISM to 61 traits and diseases from UK Biobank, we found that direct effects accounted for less than 13% of significant effects, yet were highly enriched in heritability. Multiple lines of evidence showed that PRISM causal networks are consistent with established biological mechanisms.