GWAS significance thresholds in large cohorts

Abstract

While the p-value threshold of 5.0 x 10-8 remains the standard for genome-wide association studies (GWAS) in humans and other species, it still needs to be updated to reflect the current era of large-scale GWAS, where tens of thousands of sample sizes are used to discover genetic associations at loci with smaller minor allele frequencies. In this study, we used a dataset of 348,501 individuals of European ancestry from the UK Biobank to determine the GWAS thresholds required for multiple testing corrections when considering rare and common variants in additive and dominant GWAS models. Additionally, we employed conditional and joint (COJO) analysis to quantify the proportion of false significant hits in the GWAS results for 72 traits in the UK Biobank when applying the traditional GWAS cut-off versus our newly proposed p-value thresholds. Overall, the results indicate that the conventional GWAS significance threshold of 5.0 x 10-8 yields a false positive rate of between 20% and 30% in GWAS studies that utilize large sample sizes and less common variants. Instead, a more stringent GWAS p-value threshold of 5.0 x 10-9 is needed when rare variants (with minor allele frequency > 0.1%) are included in the association test for both additive and dominance models within the European ancestry population.