Fig. 6: Performance comparison of alternative methods for prediction of two anthropometric traits (AoU-training and UKBB-tuning/validation).

We analyzed two anthropometric traits, a BMI and b height. PRS are trained using AoU data that are available for three populations: African, Latino/Admixed American, and European and then tuned in individuals from UKBB of the corresponding ancestry: AFR, AMR, and EUR (see “Real data analysis” under “Methods” for ancestry composition). Performance is reported based on adjusted R² accounting for sex, age, PC1-10 in a held-out validation sample of individuals from UKBB of the corresponding ancestry. Sample sizes for training, tuning and validation data are in Supplementary Data 7 and 8. Results for AMR are not included due to the small sample size of genetically inferred AMR ancestry individuals in UKBB. The number of SNPs analyzed in AoU analyses is much smaller than other analyses because the GWAS from AoU is on array data only (see Supplementary Data 7 for the number of SNPs). The PRS-CSx package is restricted to SNPs from HM3, whereas other alternative methods use SNPs from either HM3 or MEGA. Bars in the figure show the performance of adjusted R² for each method in each dataset. Colors are described on the right side of the figure. Source data are provided in Supplementary Data 13.