Fig. 1

Forward and reverse regression for trans-eQTL discovery. Trans-eQTLs affect multiple genes simultaneously by exerting a cis-effect on a diffusible trans-acting factor such as a transcription factor (TF) (left). In forward regression (FR), we perform univariate regression of the expression level of each gene individually on the candidate SNP’s genotype (= centered minor allele frequency) and estimate whether the distribution of resulting association p values is enriched near zero. In reverse regression (Tejaas), we perform L₂-regularized multiple regression of the candidate SNP’s genotype jointly on all gene expression levels. Crucially, reverse regression is not negatively affected by correlations between gene expression levels