Given the scale and correlations among this dataset, we found that both of these approaches provide suitable alternatives to the Bonferroni P value threshold, which in this case would be P < 1.2 × 10−11. Prioritizing the results of the permutation-based approach because it captures the data structure across all our models, we define a conservative study-wide significance cut-off of P ≤ 2 × 10−9 for the non-synonymous collapsing analysis results presented in this paper (Supplementary Tables 20, 21). Under this conservative threshold, no positive associations are expected under the null for collapsing analyses.
