A general framework for functionally informed set-based analysis: Application to a large-scale colorectal cancer study.

Author summary We developed summary statistics-based mixed effects score test statistics (sMiST) for testing the association of multiple genetically predicted mediators simultaneously and direct association of individual variants independent of mediators by using a random effects model. Extensive simulation and real data analyses demonstrate that sMiST recovers the results of MiST that is based on individual level data, but is computationally much faster. We applied our approach to a genome-wide association study of colorectal cancer and gene expression and identified several novel and secondary genetic loci. Genome-wide association studies (GWAS) have successfully identified tens of thousands of genetic variants associated with various phenotypes, but together they explain only a fraction of heritability, suggesting many variants have yet to be discovered. Recently it has been recognized that incorporating functional information of genetic variants can improve power for identifying novel loci. For example, S-PrediXcan and TWAS tested the association of predicted gene expression with phenotypes based on GWAS summary statistics by leveraging the information on genetic regulation of gene expression and found many novel loci. However, as genetic variants may have effects on more than one gene and through different mechanisms, these methods likely only capture part of the total effects of these variants. In this paper, we propose a summary statistics-based mixed effects score test (sMiST) that tests for the total effect of both the effect of the mediator by imputing genetically predicted gene expression, like S-PrediXcan and TWAS, and the direct effects of individual variants. It allows for multiple functional annotations and multiple genetically predicted mediators. It can also perform conditional association analysis while adjusting for other genetic variants (e.g., known loci for the phenotype). Extensive simulation and real data analyses demonstrate that sMiST yields p-values that agree well with those obtained from individual level data but with substantively improved computational speed. Importantly, a broad application of sMiST to GWAS is possible, as only summary statistics of genetic variant associations are required. We apply sMiST to a large-scale GWAS of colorectal cancer using summary statistics from similar to 120, 000 study participants and gene expression data from the Genotype-Tissue Expression (GTEx) project. We identify several novel and secondary independent genetic loci.