Integrating network annotation from multiple correlated traits to improve polygenic risk scores based on GWAS summary statistics

Polygenic risk scores (PRS) are valuable tools for predicting disease risk based on genetic information, with potential impacts on disease prevention and early treatment strategies. Although thousands of disease-associated genetic variants have been identified through genome-wide association studies (GWAS), the accuracy of genetic risk prediction for most diseases remains moderate and challenging. In this paper, we introduce NetPRS, a novel method that utilizes a penalized regression model and leverages network annotation information to enhance PRS prediction. This network annotation is obtained from a genotype-phenotype bipartite network (GPN), where multiple SNPs and traits are linked based on association strengths obtained from GWAS summary statistics. The network annotation allows for the incorporation of information from relevant traits into the PRS prediction for the target trait. Compared to state-of-the-art risk prediction methods, NetPRS consistently achieves improved prediction accuracy in both simulation studies and real data analysis.