Estimation of variance components for body weight of Merino sheep at birth and weaning using single nucleotide markers and REML and Bayesian approaches

Accurate estimation of variance components using pedigree and genomic data is important for prediction of breeding values. Availability of high density single nucleotide polymorphisms (SNP) arrays and genotyping many individuals resulted in the increase of accuracy of population-based estimates. Genomic selection potentially can explain all genetic variance by markers. The aim of this study was to determine the amount of genomic additive variation for birth weight and weaning weight of Merino sheep by different minor allele frequency (MAF) groups using two statistical approaches i.e. REML and Bayesian.
Material and In this study, data of 2189 Merino sheep, genotyped by 50k Illumina SNP chip were used. After the quality control of genotyping data, 47342 markers remained for subsequent analysis. For birth weight and weaning weight 1331 and 2136 records were available, respectively. To study the association between allele frequency spectrum and captured additive genetic variance, all SNPs were partitioned in five MAF bins with the equal numbers of SNPs (0-0.18, 0.18-0.28, 0.28-0.36, 0.36-0.43 and 0.43-0.499). The analysis was performed using REML and a Bayesian method implemented via Gibbs sampling and RKHS model.
Using all common SNPs in REML approach, estimates of genomic heritability were 0.58±0.07 and 0.46±0.05 for birth weight and weaning weight, respectively. This heritability values in Bayesian analysis and RKHS method for these traits were estimated to be 0.58±0.07 and 0.46 ± 0.05, respectively. The amount of genetic variation explained by five different MAF groups was different in separate and joint analysis. The estimates from separate analysis were higher than joint analysis for two traits and two (REML and Bayesian) methods. In separate analysis, the genomic heritability was similar for all MAF bins by two approaches. In the joint analysis there were large differences between REML and Bayesian estimates in terms of explaining genetic variation across MAF subsets. For birth weight, SNPs with MAF 0.18-0.28 marked the largest amount of genomic heritability using REML method. In this method, heritability of Weaning weight was zero using SNP with MAF 0.43-0.499. All MAF bins contributed to genetic variation in Bayesian method. In separate analysis, sum of genomic variances for five MAF bins was larger than estimated variance by all set of SNPs together, while sum of this variances in joint analysis was same as the value of variance obtained by all SNPs for both traits and both statistical methods.
Variance components obtained by Bayesian method are more realistic, because they have less prediction error variance. Bayesian methods considered common prior distribution for the variance, so results of this method were more reliable than other methods. Although the number of SNPs in different groups was similar, the amount of genetic variance explained by the different MAF groups was different.