Genetic factor analysis for characterizing phenome-wide patterns of genetic pleiotropy

Abstract

Genetic associations shared by multiple traits provide evidence about the biological role of disease associated variants. Here we propose genetic factor analysis (GFA), a multiphenotype analysis method that identifies common patterns of cross-trait associations, the signatures of shared biological processes. GFA overcomes limitations of alternative methods by automatically selecting the number of factors, accounting for sample overlap and allowing factors to be non-orthogonal. We apply GFA to analysis of 22 common risk factors for coronary artery disease and type 2 diabetes to partition the heritability of these traits into 15 pleiotropic components. In addition, we use GFA to obtain a biologically meaningful decomposition of a large set of blood cell composition phenotypes and use these results to increase precision in multivariable Mendelian randomization. In this application, we find that accounting for overlapping samples is critical to obtaining biologically interpretable results. Genetic factor analysis uses genome-wide association summary statistics to identify shared polygenic factors underlying patterns of cross-trait genetic associations, while automatically determining the number of factors and accounting for sample overlap.