Efficient multivariate linear mixed model algorithms for genome-wide association studies

摘要

mvLMMs are statistical regression models that relate explanatory variables to multiple correlated outcome variables and have been widely applied in genetics because of their ability to account for relatedness among samples~1. For example, they have been used to estimate the heritability of gene expression across tissues~2, assess pleiotropy and genetic correlation between complex phenotypes~(3- 6), detect quantitative trait loci7, understand evolutionary patterns8 and assist animal-breeding programs~9. Recently, mvLMMs have become increasingly important in genome-wide association studies (GWAS), both because of their effectiveness in accounting for sample relatedness~(3,7,10) and population stratification~(3,11-17), and because of a growing appreciation of the power gains of multivariate association analyses over standard univariate analysis~(3,18-22). Multivariate analyses can increase power not only to detect pleiotropic genetic variants but also genetic variants that affect only one of multiple correlated phenotypes~(22).