Our large study offers the opportunity to quantify empirically how much increasing GWAS sample sizes and ancestry diversity affects the discovery of variants, genes and biological pathways. To address this question, we re-analysed three previously published GWASs of height3,19,20 and also down-sampled our meta-analysis into four subsets (including our EUR and METAFE GWASs). Altogether, we analysed seven GWASs with a sample size increasing from around 0.13 million up to around 5.3 million individuals (Table 2).Table 2 Overview of five European-ancestry GWASs re-analysed in our study to quantify the relationship between sample size and discoveryDown-sampled GWASMax n (mean n)Number of GWS COJO SNPsPercentage of the genome covered by GWS loci (35 kb) (%)Lango Allen et al. (2010)19a130,010 (128,942)2400.5Wood et al. (2014)20241,724 (239,227)6331.4Yengo et al. (2018)3695,648 (688,927)2,7945.8GIANT-EUR (no 23andMe)1,632,839 (1,502,499)4,8679.723andMe-EUR2,502,262 (2,498,336)7,02013.6Summary statistics from the three published GWASs were imputed using the ImpG-Summary software to maximize the coverage of HM3 SNPs (Methods). GWS loci are defined as in the legend of Table 1.aSummary statistics from the Lango Allen et al. study19, initially over-corrected for population stratification using a double genomic control correction, were re-inflated such that the LD score regression intercept estimated from re-inflated test statistics equals 1.
