Genomic GxE approaches modelling heterogeneous SNP variances: applied to simulated data

Abstract

Genotype by environment interaction (GxE) can be modelled using a multi-trait approach where the same trait measured in different environments is considered a different, but correlated trait. An alternative is to model GxE with reaction norm models where the breeding values are modeled as a function of the environment defined as a continuous variable. Genomic implementations of both models can be parameterized such that  homogeneous (co)variances are assumed for all SNP across the genome. Since specific regions in the genome may harbor QTL and others may not or loci may have a large effect in one environment and a zero effect in another, the assumption of equal (co)variances across the genome is violated. We have developed an analysis protocol based on readily available BLUP software packages to allow for heterogeneous SNP (co)variances in genomic GxE models. The analysis protocol consists of a two-step approach, where the data set of interest is split in two subsets. One subset is used to estimate SNP effects and derive weights for each SNP, which are subsequently used to upweight SNP in the analysis of the second subset. We have carried out a simulation study that showed a small increase in accuracy of genomic breeding values when allowing for heterogeneous SNP (co)variances compared to homogeneous SNP (co)variances.