Genomic Prediction Using Multi-trait Weighted GBLUP Accounting for Heterogeneous Variances and Covariances Across the Genome

Abstract

Implicit assumption of common (co)variance for all loci in multi-trait Genomic Best Linear Unbiased Prediction (GBLUP) results in a genomic relationship matrix (G) that is common to all traits. When this assumption is violated, Bayesian whole genome regression methods may be superior to GBLUP by accounting for unequal (co)variance for all loci or genome regions. This study aimed to develop a strategy to improve the accuracy of GBLUP for multi-trait genomic prediction, using (co)variance estimates of SNP effects from Bayesian whole genome regression methods. Five generations (G1-G5, test populations) of genotype data were available by simulations based on data of 2,200 Danish Holstein cows (G0, reference population). Two correlated traits with heritabilities of 0.1 or 0.4, and a genetic correlation of 0.45 were generated. First, SNP effects and breeding values were estimated using BayesAS method, assuming (co)variance was the same for SNPs within a genome region, and different between regions. Region size was set as one SNP, 100 SNPs, a whole chromosome or whole genome. Second, posterior (co)variances of SNP effects were used to weight SNPs in construction of G matrices. In general, region size of 100 SNPs led to highest prediction accuracies using BayesAS, and wGBLUP outperformed GBLUP at this region size. Our results suggest that when genetic architectures of traits favor Bayesian methods, the accuracy of multi-trait GBLUP can be as high as the Bayesian method if SNPs are weighted by the Bayesian posterior (co)variances.

Multi-trait single-step genomic prediction accounting for heterogeneous (co)variances over the genome

“…The genotype data were simulated for five generations (Gen1−Gen5) based on real haplotypes of 2200 Holsteins (Gen0), as described in Karaman et al (2018). At each generation, the number of males and females were kept constant at 200 and 2000, respectively, and the mating ratio was 1:10.…”

“…Phenotypic values of the two traits were simulated to have heritabilities of 0.1 and 0.4, which represents low (L) and high (H) heritability traits, respectively. Total number of quantitative trait loci (QTL) was set at 200, which were randomly selected from the SNP set, ensuring that the average minor allele frequency (MAF) of QTL is 0.15 (Karaman et al 2018). The criterion for the MAF of the QTL was based on the assumption that they in general have relatively low MAF (Goddard and Hayes 2009;Kemper and Goddard 2012).…”

“…Although it ignores the fact that a proportion of the genome regions, and therefore a proportion of the SNPs, may not have any influence on the trait of interest, prediction accuracy may increase compared to BayesA by benefiting from the increase in the accuracy in estimation of SNP variances, and compared to BLUP-based models by allowing SNPs in different regions to have different variances. Partitioning of the covariate matrix of marker genotypes, M, or in other words, assigning priors to genome regions rather than individual SNPs, was shown to influence the accuracy of genomic predictions (Brøndum et al 2012;Gebreyesus et al 2017;Karaman et al 2018).…”

“…Many important traits in animal breeding have genetic correlations in varying sizes with one or more traits, and therefore, measurements of such correlated traits carry information for the genetic values of others. Several multitrait models have been proposed for genomic prediction (Calus and Veerkamp 2011;Jia and Jannink 2012;Hayashi and Iwata 2013;Gebreyesus et al 2017;Cheng et al 2018b), and simulations have shown that genomic prediction accuracies from multi-trait models are superior to those from single-trait models (Calus and Veerkamp 2011;Jia and Jannink 2012;Guo et al 2014;Karaman et al 2018). Multitrait genetic evaluation rely on the genetic association between the traits through the genetic variance and covariance structure.…”

“…Models used for genomic prediction, therefore, should properly account for the makeup of these genetic (co)variance components to obtain the highest accuracy of prediction. When only a few genome regions explain a considerable amount of the variances and/or covariance in a two-trait analysis, models that account for the heterogeneous correlation structure over the genome may have advantages over the methods that assumes a constant correlation over the genome (Gebreyesus et al 2017;Karaman et al 2018).…”