eQTL Identification
An overview of our workflow can be found in Figure S1. First, to identify primary and conditional cis-eQTL, we a conducted forward stepwise conditional analysis implemented in MatrixEQTL21 using genotype data at 6.4 million markers and RNA-seq data for 16,423 genes. FDR was initially assessed using the Benjamini-Hochberg algorithm across all cis-eQTL tests within each chromosome. FDR was not re-assessed at each conditional step; instead, a fixed p value threshold was used as the inclusion criteria in the stepwise model selection. For each gene with at least one cis-eQTL (gene ± 1 Mb) association at a 5% false discovery rate (FDR), the most significant SNP was added as a covariate in order to identify additional independent associations (considered significant if the p value achieved was less than that corresponding to the initial 5% FDR for primary eQTL). This procedure was repeated iteratively until no further eQTL met the p value threshold criteria. We used a linear regression model, adjusting for diagnosis and five ancestry covariates inferred by GemTools. Following eQTL identification, only autosomal eQTL were retained for downstream analyses.