Integrating multiple traits for improving polygenic risk prediction in disease and pharmacogenomics GWAS.

This article focuses on the development and evaluation of novel multi-trait polygenic risk score (mtPRS) methods for improving genetic risk prediction and association testing by integrating information from multiple genetically correlated traits. The authors propose mtPRS-PCA, which constructs a composite PRS by weighting single-trait PRSs using principal component analysis (PCA) on the genetic correlation matrix, and mtPRS-O, an omnibus method that combines association test P values from mtPRS-PCA, machine-learning-based mtPRS (mtPRS-ML), and single-trait PRSs using the Cauchy Combination Test for robust inference. Extensive simulations under diverse genetic architectures demonstrate that mtPRS-PCA outperforms existing methods when traits have similar correlations, dense effect signals, and concordant effect directions, while mtPRS-O provides robust association testing across scenarios. Application to pharmacogenomics genome-wide association study (GWAS) data from a randomized clinical trial of anacetrapib shows that mtPRS-PCA improves prediction accuracy and patient stratification for low-density lipoprotein cholesterol response, and mtPRS-O maintains robust association performance, supporting their utility in both disease and drug response genetic studies.