Genomic prediction of breeding values in the New Zealand sheep industry using a 50K SNP chip
Genomic prediction of breeding values in the New Zealand sheep industry using a 50K SNP chip
复制标题
DOI：
10.2527/jas.2014-7801
复制
发表时间：
2014-10-01
影响因子：
3.3
通讯作者：
Dodds, K. G.
中科院分区：
文献类型：
--
作者：
Auvray, B.;McEwan, J. C.;Dodds, K. G.
基因组预测的目的是从基因组数据中预测育种价值。我们描述了用于工业用途的分子育种值(MBV)的基因组预测方程和精度的发展，重点介绍了用于处理新西兰绵羊种群结构预测的方法。这是由纯种和杂交动物组成的混合体，但主要以罗姆尼为基础。特别是，我们使用基于系谱的EBV对8个性状(断奶体重作为直接效应，断奶体重作为母体效应，8月龄活重，12月龄活重，12月龄油毛重，羔羊体重，成年羊毛重，产羔数)和来自13,420只动物的Illumina OvineSNP50珠芯片基因型进行了基于SNP标记的BLUP与不同基因组关系矩阵(GRM)的研究，并调查了不同年龄组动物(训练集)以预测年轻动物的MBV(验证集)。测试的GRM包括修改以解决品种之间的等位基因频率差异，重新缩放以使GRM的平均值等于传统的系谱分子关系矩阵A的平均值，以及使用凸组合将GRM与A相结合，该组合具有通过最大化条件限制似然估计的权重。我们发现这些修改是有益的，并建议使用品种调整的GRM与A训练数据集相结合，这些训练数据集与Romney、Coopworth和Perendale动物一起使用通常比只使用一个纯粹的品种训练数据集对所有性状的预测更好。但对佩伦代尔品种的预测更准确，因为佩伦代尔训练集对8个性状中的3个进行了训练。我们的结论是，对所有性状和品种的组合使用混合品种训练集是最好的，但建议增加Perendale动物的基因分型数量，以提高该品种的MBV准确性。
The aim of genomic prediction is to predict breeding value from genomic data. We describe the development of genomic prediction equations and accuracies for molecular breeding values (MBV) for industry use, focusing on the methodology used to deal with predictions for the New Zealand sheep population structure. This is made up of a mixture of pure and crossbred animals, but principally Romney based. In particular, we used pedigree-based EBV for 8 traits (weaning weight as a direct effect, weaning weight as a maternal effect, live weight at 8 mo, live weight at 12 mo, greasy fleece weight at 12 mo, lamb fleece weight, adult fleece weight, and number of lambs born) and Illumina OvineSNP50 BeadChip genotypes from 13,420 animals to investigate BLUP with different genomic relationship matrices (GRM) based on SNP markers and to investigate varying sets of older animals (training sets) to predict the MBV of younger animals (validation sets). The GRM tested included modifications to account for allele frequency differences between breeds, rescaling so that the mean GRM is equal to the mean of the traditional pedigree numerator relationship matrix A, and combining of the GRM with A using a convex combination with a weight estimated by maximizing a conditional restricted likelihood. We found that these modifications were beneficial and recommend using a breed-adjusted GRM combined with A. Training data sets with Romney, Coopworth, and Perendale animals all together usually predicted better than using just a pure breed training data set for all traits. But predictions for the breed Perendale were more accurate with a Perendale training set for 3 of the 8 traits. We concluded that using a mixed-breed training set for all combinations of traits and breeds was best but advise that increasing the number of Perendale animals genotyped should be a priority to increase the MBV accuracies obtained for that breed.