Fig. 5
From: Optimization of multi-ancestry polygenic risk score disease prediction models
Predictive performance of risk models that incorporate ancestry, age, sex and risk factor information. (A) Prediction accuracy was measured using AUC in the UKB testing set. Bars indicate 95% confidence intervals of 10,000 non-parametric bootstrap replicates. Models shown include performance with: ensemble PRS alone; the addition of ancestry PCs, ancestry with Orchestra for local ancestry deconvolution in addition to PCs; sex and age in addition to ancestry, and relevant risk factors in addition to all other information. (B) Prediction results summarized per risk model where each dot represents a trait. If a trait lacked relevant risk factors, the accuracy of the final comprehensive model equaled that of the preceding model, which included only prs, ancestry, age and sex. (C) Average weights per feature in the final logistic regression model across 30 studied traits.
