Almost Free Enhancement of Multi-Population PRS: From Data-Fission to Pseudo-GWAS Subsampling

Area of Science:

• Genetics
• Biostatistics
• Population Health

Background:

• Multiple polygenic risk score (PRS) methods exist to enhance prediction accuracy in diverse populations.
• No single PRS method demonstrates consistent superiority across all data scenarios.
• Integrating PRS results across methods and populations can improve predictions but often requires individual-level data for weight calculation, which is not always available.

Purpose of the Study:

• To introduce MIXPRS, a robust framework for integrating multi-population PRS methods.
• To enable effective combination of PRS methods using only genome-wide association study (GWAS) summary statistics.
• To improve PRS prediction accuracy, particularly in underrepresented groups.

Main Methods:

• MIXPRS utilizes data fission principles for PRS integration.
• Applies SNP pruning to address linkage disequilibrium (LD) mismatch between GWAS summary statistics and LD reference panels.
• Employs non-negative least squares regression for robust estimation of PRS combination weights.

Main Results:

• MIXPRS consistently outperforms existing methods in prediction accuracy across diverse traits and populations.
• Demonstrated effectiveness in simulations and real-data analyses using UK Biobank and All of Us datasets.
• Achieved superior performance for 22 continuous and four binary traits.

Conclusions:

• MIXPRS offers a broadly accessible, robust, and generalizable solution for PRS integration.
• The framework's reliance solely on GWAS summary statistics overcomes limitations of previous methods.
• Enhances the utility of PRS for improving health predictions in underrepresented populations.