Polygenic risk scores: effect estimation and model optimization

Background : Polygenic risk score (PRS) derived from summary statistics of genome-wide association studies (GWAS) is a useful tool to infer an individual ’ s genetic risk for health outcomes and has gained increasing popularity in human genetics research. PRS in its simplest form enjoys both computational ef fi ciency and easy accessibility, yet the predictive performance of PRS remains moderate for diseases and traits. Results : We provide an overview of recent advances in statistical methods to improve PRS ’ s performance by incorporating information from linkage disequilibrium, functional annotation, and pleiotropy. We also introduce model validation methods that fi ne-tune PRS using GWAS summary statistics. Conclusion : In this review, we showcase methodological advances and current limitations of PRS, and discuss several emerging issues in risk prediction research. Author summary: The prosperity of powerful genome-wide association studies (GWASs) has facilitated rapid development of polygenic risk score (PRS). Many post-GWAS PRS methods have been introduced to directly address the mediocre prediction accuracy of traditional PRS built upon marginal estimates from GWAS. This review fi rst summarizes PRS methods inspired by different biological concepts including LD, functional annotation, and pleiotropy to better quantify SNP effects. Then we introduce recent PRS frameworks that enable model optimization using summary statistics. Finally, we point out current pitfalls of risk prediction research. We expect emerging methods that address current challenges in the near future.