False discovery rates for genome-wide association tests in biobanks with thousands of phenotypes

Biobanks housing genetic and phenotypic data for thousands of individuals introduce new opportunities and challenges for genetic association studies. Association testing across many phenotypes increases the multiple-testing burden and correlation between phenotypes makes appropriate multiple-testing correction uncertain. Moreover, analysis including low-frequency variants results in inflated type I error due to the much larger number of tests and the elevated importance of each individual minor allele carrier in those tests. Here we demonstrate that standard Bonferroni and permutation-based methods for multiple testing correction are inadequate for a holistic analysis of biobank data because ideal significance thresholds vary across datasets and minor allele frequencies. We propose a single-iteration permutation method that is computationally feasible and provides false discovery rate (FDR) estimates tailored to individual datasets and variant frequencies. Each dataset’s unique FDR estimates provide customized levels of confidence for association results and enable informed interpretation of genetic association studies across the phenome.