One score to rule them all: regularized ensemble polygenic risk prediction with GWAS summary statistics

Abstract

Ensemble learning has become a cornerstone for improving the predictive accuracy of polygenic risk scores (PRS), and nearly all recent multi-ancestry PRS methods incorporate ensemble learning as a final step. However, existing ensemble approaches require individual-level genotype data for model training, which limits their real-world applications, especially in non-European populations without sufficient genomic samples. Here, we introduce a statistical framework for constructing regularized ensemble PRS that integrates a large number of candidate PRS models using only genome-wide association study summary statistics. Through extensive analyses across multiple traits and populations, we demonstrate that our method consistently outperforms state-of-the-art PRS approaches within and across ancestries. This framework presents "one score to rule them all" for its capability to enable seamless integration of newly developed PRS models with existing ones, providing a scalable and general solution for future PRS development and application.