Improving on polygenic scores across complex traits using select and shrink with summary statistics

Abstract

Structured AbstractO_ST_ABSMotivationC_ST_ABSAs precision medicine advances, polygenic scores (PGS) have become increasingly important for clinical risk assessment. Many methods have been developed to create polygenic models with increased accuracy for risk prediction. Our select and shrink with summary statistics (S4) PGS method extends a previous method (polygenic risk score - continuous shrinkage (PRS-CS)) by using a continuous shrinkage prior on effect sizes with a selection strategy for including SNPs to create the best performing model. ResultsThe S4 method provides overall improved PGS accuracy for UK Biobank participants when compared to LDpred2 and PRS-CS across a variety of phenotypes with differing genetic architectures. Additionally, the S4 method has higher estimated PGS accuracy over LDpred2 in Finnish and Japanese populations. Thus, the S4 method represents an improvement in overall PGS accuracy across multiple phenotypes and increases the transferability of PGS across ancestries. Availability and ImplementationThe S4 program is freely available at https://github.com/jpt34/S4_programs. Supplementary informationSupplementary data [will be] available at Bioinformatics online.