A new method for multiancestry polygenic prediction improves performance across diverse populations.

Area of Science:

• Genetics and Genomics
• Population Health
• Bioinformatics

Background:

• Polygenic risk scores (PRS) are valuable for predicting complex traits but show reduced performance in non-European populations.
• This performance gap raises concerns regarding clinical applications and exacerbates health inequities.
• Existing methods for PRS calculation may not adequately address ancestral diversity.

Purpose of the Study:

• To develop and evaluate CT-SLEB, a novel and scalable method for calculating polygenic risk scores.
• To improve PRS performance specifically in non-European populations.
• To assess the impact of sample size and single nucleotide polymorphism (SNP) density on multiancestry risk prediction.

Main Methods:

• Developed CT-SLEB, integrating clumping and thresholding, empirical Bayes, and superlearning.
• Utilized ancestry-specific genome-wide association study (GWAS) summary statistics from multiancestry training samples.
• Evaluated CT-SLEB against nine alternative methods using large-scale simulated GWAS data and real-world datasets (23andMe, GLGC, All of Us, UK Biobank) involving over 5.1 million individuals.

Main Results:

• CT-SLEB significantly enhanced PRS performance in non-European populations compared to simpler methods.
• CT-SLEB demonstrated comparable or superior performance to a recent, computationally intensive method.
• Simulation studies provided insights into optimal sample size and SNP density for multiancestry PRS.

Conclusions:

• CT-SLEB offers a powerful and scalable solution for improving PRS accuracy across diverse ancestries.
• The developed method has the potential to reduce health disparities associated with PRS.
• Findings underscore the importance of ancestry-specific approaches in genetic risk prediction.