To learn the mixture weights and scaling for the main and interacting effects, we initially fitted the mash model with a random set (that is, unbiased representation of the results) of the nominal eQTL results (that is, 5% for gene–variant pairs and 1% for transcript–variant, exon–variant and junction–variant pairs). We next fitted these mixture weights and scaling to all of the main and interacting eQTL results in chunks. After model fitting, we extracted posterior summaries and measures of significance (that is, the LFSR). We considered main and interacting eQTLs significant if the LFSR < 0.05.
