CORE GREML: Estimating covariance between random effects in linear mixed models for genomic analyses of complex traits

Abstract

Linear mixed models (LMMs) using genome-based restricted maximum likelihood (GREML) are a key variance partitioning tool, where effects of multiple sources, such as different functional genomic regions, on phenotypes are treated as random. Classic LMMs assume independence between random effects, which can cause biased estimation of variance components. Here, we relax this independence assumption by introducing a generalised GREML, called CORE GREML, that can explicitly estimate the covariance between random effects. Using extensive simulations, we show that CORE GREML outperforms the conventional GREML, providing unbiased estimates of variance and covariance components. Using data from the UK biobank, we demonstrate that CORE GREML is useful for genomic partitioning analyses and for genome-transcriptome partitioning of phenotypic variance. For example, we found that the transcriptome, imputed using genotype data, explained a significant proportion of phenotypic variance for height (0.15, se = 5.4e-3, p-value = 1.5e-283), and that these transcriptomic effects on phenotypes correlated with effects of the genome (r = 0.35, se = 4.6e-2, p-value = 1.2e-14). We conclude that the covariance between random effects is a key parameter that needs to be estimated, especially when partitioning phenotypic variance by omic layer.