Correction for multiple comparisons must also be applied to pathway p-values to control for false positives. As in other areas of statistical genomics, optimizing methods for correction is a work in progress. Bonferroni-related methods seem too conservative for pathway analyses because they do not allow for dependence across pathways. False discovery rate (FDR) approaches [64] are frequently-applied in pathway analyses [6, 26, 48], while newer FDR-based [65] and bootstrapping [39] methods that permute on raw genotypes can better account for pathway overlap but require large computational capacity.
