Leveraging fine-mapping and multipopulation training data to improve cross-population polygenic risk scores

Area of Science:

• Genetics
• Bioinformatics
• Population Health

Background:

• Polygenic risk scores (PRS) demonstrate reduced accuracy in non-European populations, contributing to health disparities.
• Existing methods struggle with differing linkage disequilibrium patterns across ancestries.

Purpose of the Study:

• To develop and evaluate PolyPred, a novel method to enhance the accuracy of cross-population polygenic risk scores.
• To address the limitations of current PRS methods in diverse ancestral groups.

Main Methods:

• PolyPred combines a new functionally informed fine-mapping predictor (estimating causal effects) with BOLT-LMM.
• PolyPred+ incorporates non-European training data when available for further improvement.
• Analogs of PolyPred and PolyPred+ were developed using summary statistics.

Main Results:

• PolyPred improved PRS accuracy by 7-32% in South Asian and African populations compared to BOLT-LMM, validated by simulations.
• PolyPred+ achieved 24% improvement over BOLT-LMM and 12% over PolyPred in East Asian populations.
• Summary statistics-based methods showed comparable performance gains.

Conclusions:

• PolyPred and PolyPred+ significantly enhance polygenic risk score accuracy across diverse populations.
• These methods offer a promising approach to mitigate health disparities stemming from PRS inaccuracies.
• The functionally informed fine-mapping component is key to addressing linkage disequilibrium differences.