Whole genome evaluation of quantitative traits using suitable statistical methods enables researchers to predict genomic breeding values (GEBVs) more accurately. Recent studies suggested that the ability of methods in terms of predictive performance may depend on the genetic architecture of traits. Therefore, when choosing a statistical method, it is essential to consider the genetic architecture of the target traits. Herein, the performance of parametric methods i.e. GBLUP and BayesB and non-parametric methods i.e. Bagging GBLUP and Random Forest (RF) were compared for traits with different genetic architecture. Three scenarios of genetic architecture, including purely Additive (Add), purely Epistasis (Epis) and Additive-Dominance-Epistasis (ADE) were considered. To this end, an animal genome composed of five chromosomes, each chromosome harboring 1000 SNPs and four QTL was simulated. Predictive accuracies in the first generation of testing set under Additive genetic architectures for GBLUP, BayesB, Baging GBLUP and RF were 0.639, 0.731, 0.633 and 0.548, respectively, and were 0.278, 0.330, 0.275 and 0.444 under purely Epistatic genetic architectures. Corresponding values for the Additive-Dominance-Epistatic structure also were 0.375, 0.448, 0.369 and 0.458, respectively. The results showed that genetic architecture has a great impact on prediction accuracy of genomic evaluation methods. When genetic architecture was purely Additive, parametric methods and Bagging GBLUP were better than RF, whereas under Epistatic and Additive-Dominance-Epistatic genetic architectures, RF delivered better predictive performance than the other statistical methods.

中文翻译：

---

使用非加性遗传结构评估用于预测数量性状的参数和非参数方法的评估

使用适当的统计方法对数量性状进行全基因组评估，使研究人员能够更准确地预测基因组育种值（GEBV）。最近的研究表明，方法在预测性能方面的能力可能取决于性状的遗传结构。因此，在选择统计方法时，必须考虑目标性状的遗传结构。在此，针对具有不同遗传结构的性状，比较了参数方法（即GBLUP和BayesB）和非参数方法（即装袋GBLUP和随机森林（RF））的性能。考虑了三种遗传结构方案，包括纯加性（Add），纯上位性（Epis）和加性-优势-表皮病（ADE）。为此，一个动物基因组由五个染色体组成，每个带有1000个SNP和4个QTL的染色体都经过模拟。在GBLUP，BayesB，Baging GBLUP和RF的加性遗传架构下，第一代测试集中的预测准确性分别为0.639、0.731、0.633和0.548，在纯上位遗传架构下分别为0.278、0.330、0.275和0.444。加性-显性-上位性结构的相应值也分别为0.375、0.448、0.369和0.458。结果表明，遗传结构对基因组评价方法的预测准确性有很大影响。当遗传架构纯粹是可加的时，参数方法和Bagging GBLUP优于RF，而在上位遗传和加性-显性-上位性遗传架构下，RF的预测性能优于其他统计方法。