The (in)famous GWAS P-value threshold revisited and updated for low-frequency variants

Abstract

Genome-wide association studies (GWAS) have long relied on proposed statistical significance thresholds to be able to differentiate true positives from false positives. Although the genome-wide significance P-value threshold of 5 × 10(-8) has become a standard for common-variant GWAS, it has not been updated to cope with the lower allele frequency spectrum used in many recent array-based GWAS studies and sequencing studies. Using a whole-genome- and -exome-sequencing data set of 2875 individuals of European ancestry from the Genetics of Type 2 Diabetes (GoT2D) project and a whole-exome-sequencing data set of 13 000 individuals from five ancestries from the GoT2D and T2D-GENES (Type 2 Diabetes Genetic Exploration by Next-generation sequencing in multi-Ethnic Samples) projects, we describe guidelines for genome- and exome-wide association P-value thresholds needed to correct for multiple testing, explaining the impact of linkage disequilibrium thresholds for distinguishing independent variants, minor allele frequency and ancestry characteristics. We emphasize the advantage of studying recent genetic isolate populations when performing rare and low-frequency genetic association analyses, as the multiple testing burden is diminished due to higher genetic homogeneity.

Figure 1

Impact of linkage disequilibrium pruning on P-value significance threshold for whole-genome and -exome association studies of European ancestries (N=2875). Variants with an LD between each other bigger or equal than a certain threshold are pruned, so only randomly chosen tagging SNPs are used for the multiple testing P-value threshold calculation. (a) GoT2D WGS integrated SNP panel with horizontal line showing the commonly used 5 × 10⁻⁸ genome-wide significance P-value threshold. (b) WES SNP panel with horizontal line showing the commonly used 5 × 10⁻⁷ exome-wide significance Bonferroni corrected P-value threshold for MAF≥0.5%. The plotted P-values were calculated as 0.05/number of tag SNPs at each described MAF and LD pruning thresholds (Materials and methods; Supplementary Table S1; Supplementary Table S2). The data points with LD>1 refer to no LD pruning.

Figure 2

P-value needed to reach WGS at different MAF and LD thresholds for a whole-genome analysis of Finnish, Swedish and British populations (a) and whole-exome-sequencing analysis of diverse ancestries (b) using the same sample sizes across ancestries (N=512 for WGS, N=861 for WES). The horizontal lines show the commonly used significance P-value threshold for WGS (5 × 10⁻⁸) and WES (5 × 10⁻⁷). Supplementary Figure S3 presents the multiple ancestry data divided into separate plots.