Optimization of Multi-Ancestry Polygenic Risk Score Disease Prediction Models

Polygenic risk scores (PRS) have ushered in a new era in genetic epidemiology, offering insights into individual predispositions to a wide range of diseases. This study aimed to develop and benchmark multi-ancestry PRS models capable of predicting disease risk across diverse populations. Leveraging trans-ethnic GWAS meta-analysis, we generated novel summary statistics for 30 medically-related traits and assessed the predictive performance of four PRS algorithms. Algorithm efficacy across traits varied, with Stacked C+T (SCT) performing better when trait prevalence was higher, while LDpred and Lassosum performed better when trait prevalence was lower. Subsequent integration of PRS algorithm outputs through logistic regression and incorporation of additional medical information enhanced model accuracy. Notably, including ancestry information further improved predictive performance, underscoring its importance in PRS model development. Our findings highlight the potential of multi-ancestry PRS models in clinical settings, demonstrating superior predictive accuracy and broad applicability across different ancestral backgrounds.