Bayesian optimization is incorporated into genomic prediction to identify the best genotype from a candidate population. Several expected improvement (EI) criteria are proposed for the Bayesian optimization. The iterative search process of the optimization consists of two main steps. First, a genomic BLUP (GBLUP) prediction model is constructed using the phenotype and genotype data of a training set. Second, an EI criterion, estimated from the resulting GBLUP model, is employed to select the individuals that are phenotyped and added to the current training set to update the GBLUP model until the sequential observed EI values are less than a stopping tolerance. Three real datasets are analyzed to illustrate the proposed approach. Furthermore, a detailed simulation study is conducted to compare the performance of the EI criteria. The simulation results show that one augmented version derived from the distribution of predicted genotypic values is able to identify the best genotype from a large candidate population with an economical training set, and it can therefore be recommended for practical use. Supplementary materials accompanying this paper appear on-line.

贝叶斯优化被并入基因组预测中，以从候选群体中识别出最佳基因型。针对贝叶斯优化，提出了几种预期的改进（EI）标准。优化的迭代搜索过程包括两个主要步骤。首先，使用训练集的表型和基因型数据构建基因组BLUP（GBLUP）预测模型。其次，从所得的GBLUP模型估算出的EI标准用于选择表型个体，并将其添加到当前训练集中以更新GBLUP模型，直到顺序观察到的EI值小于终止耐受性为止。分析了三个真实的数据集以说明所提出的方法。此外，进行了详细的模拟研究以比较EI标准的性能。模拟结果表明，从预测的基因型值分布中获得的一种增强版本能够通过经济的训练集从大量候选人群中识别出最佳基因型，因此可以推荐用于实际应用。本文随附的补充材料在线出现。