Across ancestry groups, the highest accuracy of PGSGWS-METAFE was observed in EUR participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{GWS}}-{\rm{METAFE}}}^{2}$$\end{document}RGWS−METAFE2~40%; s.e. 0.6%) and the lowest in AFR participants from the UKB (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{GWS}}-{\rm{METAFE}}}^{2}$$\end{document}RGWS−METAFE2 ≈ 9.4%; s.e. 0.7%). Note that the difference in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{GWS}}-{\rm{METAFE}}}^{2}$$\end{document}RGWS−METAFE2 between the EUR and AFR ancestry cohorts is expected because of the over-representation of EUR in our METAFE, and consistent with a relative accuracy (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{GWS}}-{\rm{METAFE}}}^{2}$$\end{document}RGWS−METAFE2 in AFR)/(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,{R}_{{\rm{GWS}}-{\rm{METAFE}}}^{2}$$\end{document}RGWS−METAFE2 in EUR) of around 25% that was previously reported30. We extended analyses of PGSGWS to PGS based on SNPs identified with COJO at lower significance thresholds (Extended Data Fig. 7). As in previous studies3,20, the inclusion of sub-significant SNPs increased the accuracy of ancestry-specific PGSs. However, lowering the significance thresholds in our METAFE mostly improved accuracy in EUR (from 40% to 42%), whereas it slightly decreased the accuracy in AFR.
