With this magnitude of test statistics generated in the PheWAS scale, another proposal for P value thresholding involves n-of-1 permutation58. In applying this approach, we shuffled the case–control (or quantitative measurement) labels once for every phenotype while maintaining the participant-genotype structure and across all 11 non-synonymous collapsing models for binary traits (3,582,997,902 tests) and quantitative traits (292,856,058 tests). Reviewing the tails of these two P value distributions, the lowest permutation-based P value achieved was 1.9 × 10−9 (binary tests) and 3.2 × 10−9 (quantitative tests).
