MIXPRS enables multi-population and multi-method polygenic risk scores using summary statistics

Many multi-population polygenic risk score (PRS) methods have been proposed to improve prediction in underrepresented populations; however, no single method performs best across all scenarios. Although integrating PRSs across multiple methods and populations may improve prediction, this approach is often limited by the need for individual-level tuning data. Here we introduce MIXPRS, a robust framework based on the data fission paradigm for combining multiple multi-population PRS methods using only genome-wide association study summary statistics. MIXPRS uses single nucleotide polymorphism pruning to mitigate linkage disequilibrium mismatch and non-negative least squares regression to estimate combination weights. Across simulations and real-data analyses spanning up to 26 traits, MIXPRS consistently improves prediction accuracy over existing methods. We further extend this framework to MIXPRS+, incorporating functional annotations and clinical PRSs, yielding additional gains in non-European populations. MIXPRS relies only on summary statistics, thus offering broad accessibility and robustness for underrepresented populations.