Extended Data Figure 3: Robustness of directional selection signals (related to Figure 1a,b).

(a) Proportion of SNPs significant in any of 452 pan-UK Biobank GWAS studies for X-statistics with magnitudes larger than the threshold on the x-axis, adjusted for minor allele frequency and measures of linked purifying selection (McVicker-B, Murphy-phastCons, and Murphy-CADD). Background selection tends to be higher in functional genomic regions, so SNPs with higher $\left|\mathrm{X}\right|$ are more penalized than in Figure 1a hence the lower plateau. (b) Simulating neutral, negative, and positive selection for a 200 kb window around a focal SNP, with derived allele frequency drawn uniformly from [0,1]. The focal SNP has $s=0.01$, population size is constant at 20000 diploid individuals, mutation rate per base pair per generation is 2×10⁻⁸, and recombination rate is 1 cM per 1 Mb. (c) Residual mean (HAF)/n for a haploid sample size n over 200 bp windows is observed minus expected value. Expected value is determined using a linear regression model with McVicker-B, Murphy-phastCons, and Murphy-CADD as variables, providing the expected mean (HAF)/n conditioned on them.