Fig. 3: Trans-ancestry relative prediction accuracy of PGS of 5 quantitative traits and three common diseases. | Nature Communications

Fig. 3: Trans-ancestry relative prediction accuracy of PGS of 5 quantitative traits and three common diseases.

From: Theoretical and empirical quantification of the accuracy of polygenic scores in ancestry divergent populations

Fig. 3

a–c Relative accuracies (RA) are calculated as the ratio of the squared correlation between PGS and traits/diseases in UKB participants of non-European ancestry over the same squared correlation estimated in ~20,000 independent UKB participants of European ancestry. We report here only ancestry-trait/disease pairs, with a significant reduction in RA (Wald test, p-value <0.05). Data for all ancestry-trait/disease pairs are provided in Supplementary Table 2. RApred(LD+MAF) refers to the RA predicted from Eq. (2) only using information from LD and MAF differences between ancestries. RAobs refers to observed RA calculated using independent genome-wide significant trait-associated SNPs. Panels d–f show the proportion of the loss of accuracy (LOA) explained by LD and MAF (Supplementary Fig. 12) calculated as 100% × (1 − RApred(LD+MAF))/(1 − RAobs). The grey dashed lines are y = 100% and y = 50%. Error bars in the figure represent the standard errors of observed RA or proportion of LOA explained by LD and MAF in each ancestry-trait/disease pair, which calculation is detailed in Supplementary Note 7.

Back to article page