JointPRS: A data-adaptive framework for multi-population genetic risk prediction incorporating genetic correlation

Area of Science:

• Genetics
• Bioinformatics
• Population Health

Background:

• Genetic risk prediction accuracy is limited in non-European populations due to smaller Genome-Wide Association Study (GWAS) sample sizes.
• Existing methods struggle with limited tuning datasets, hindering effective polygenic risk score (PRS) development for diverse ancestries.

Purpose of the Study:

• To introduce JointPRS, a novel data-adaptive framework designed to improve genetic risk prediction across multiple populations.
• To enable accurate PRS development without requiring individual-level tuning data, even with small tuning sets.

Main Methods:

• JointPRS leverages genetic correlations across populations using GWAS summary statistics.
• The framework employs a data-adaptive approach to optimize predictions.
• Performance was evaluated through extensive simulations and real-world applications using UK Biobank (UKBB) and All of Us (AoU) data.

Main Results:

• JointPRS consistently outperformed six state-of-the-art methods across various data scenarios (no tuning, same-cohort, cross-cohort).
• Significant improvements were observed in the Admixed American population, with lipid trait prediction accuracy increasing by 6.46%-172.00% in the AoU cohort.
• The method demonstrated effectiveness across 22 quantitative and four binary traits.

Conclusions:

• JointPRS offers a robust solution for enhancing genetic risk prediction in diverse populations, addressing current limitations in GWAS and tuning data availability.
• The framework's data-adaptive nature and ability to utilize cross-population genetic correlations provide substantial improvements, particularly for underrepresented groups.