Overall, ancestry-specific PGSHM3 consistently outperform their corresponding PGSGWS in most ancestry-groups. However, PGSHM3 was sometimes less transferable across ancestry groups than PGSGWS, in particular in AFR and HIS individuals from PAGE. In EUR, PGSHM3 reaches an accuracy of 44.7% (s.e. 0.6%), which is higher than previously published SNP-based predictors of height derived from individual-level data31–33 and from GWAS summary statistics28,34,35 across various experimental designs (different SNP sets, different sample sizes and so on). Finally, the largest improvement of PGSHM3 over PGSGWS was observed in AFR individuals from the PAGE study (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{GWS}}-{\rm{AFR}}}^{2}$$\end{document}RGWS−AFR2 = 8.5% versus \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{HM}}3}^{2}$$\end{document}RHM32 = 15.4%; Fig. 4a) and the UKB (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{GWS}}-{\rm{AFR}}}^{2}$$\end{document}RGWS−AFR2 = 8.5% versus \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{{\rm{HM}}3}^{2}$$\end{document}RHM32 = 14.4%; Fig. 4a).
