Despite this monotonically decreasing trend on average across traits, we found the observed RA of the LDL PGS to be larger in participants of EAS (RA = 58%; standard error S.E. 10%) and AFR (RA = 40%; standard error S.E. 5.0%) ancestries as compared with participants of SAS ancestry (RA = 35%; S.E. 4.1%). We further investigated this observation and found this pattern to be explained by a single large effect variant (namely rs7254892 with an estimated SNP effect \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat \beta$$\end{document}β^ = 0.49 standard deviation per allele, MAF estimated in 1KGP = 0.03, 0.04, 0.07 and 0.16 in EUR, SAS, EAS and AFR ancestries, respectively) which explains 1.5% of LDL variance in participants of EUR ancestry. Excluding this variant from the LDL PGS led to a reduced accuracy in participants of EAS and AFR ancestries (RA = 30%; S.E. 7.8% for EAS and RA = 16%; S.E. 3.3% for AFR; Supplementary Fig. 11) below that from participants of SAS ancestry (RA = 37%; S.E. 4.5%). This result is important as it
