By contrast, pathway enrichment tools assess for a statistically-significant distribution of association within a pathway. Competitive enrichment methods compare the collective association within a pathway to the collective signal among genes not in the pathway [45]. As a result, competitive methods are not suitable for candidate pathway analyses that do not have an appropriate complement of data from outside of the candidate pathways. Meanwhile, self-contained enrichment methods test the signal within a pathway against simulated data sets which are expected to have no significant phenotype association [45, 46]. Self-contained methods can be challenging to use in a screening-oriented GWPA due to the computational demand of generating simulated data sets. In addition, self-contained approaches are particularly susceptible to false positives through genomic inflation, as each pathway is evaluated independently from any other data on the source assay. While one study [47] normalized all association statistics to a genomic inflation factor calculated by PLINK, best practices in this area have not yet been settled. Competitive tests are more robust in controlling genomic inflation, but they can also relinquish power in data sets with diffuse association signal [45]. As such, the optimal method depends on study goals, data set properties, and computational resources.
