Extended Data Fig. 4: Calibration of single-variant analysis.

To assess whether our single-variant association statistics (two-sided, calculated by the EMMAX test) were well-calibrated, we computed quantile–quantile plots of associations across all samples (Overall) and within each ancestry (total n = 45,231 individuals). To avoid deflation of the quantile–quantile plot from rare variants (for which the expected P values are discrete rather than uniformly distributed), only variants with minor allele counts of 20 or greater (either overall or within the relevant ancestry) are shown. Variants were also LD-pruned before plotting, to avoid induced variance from correlated P values of these variants, using the ‘clump’ method implemented in PLINK. The λ values indicate genomic control, as measured by the ratio in observed median χ² statistic to that expected under the null hypothesis. Red line, expectation of P values under the null distribution. Blue lines (and grey region), 95% confidence interval of expectations under the null distribution.