用於從候選族群中選拔最佳基因型A-最適與 D-最適訓練集之研究

Abstract

隨著分子生物學的進步，基因體選拔 (genomic selection，GS)廣泛用於動物或作物育種計畫中，並成為一項重要的工具。儘管基因型分析 (genotyping)的成本降低，外表型分析 (phenotyping)仍然是要花相對較高的成本以及時間，因此希望透過基因型 (genotype)推測外表型(phenotype)，以此加速育種計畫。基因體選拔透過遍布整個基因體 (genome)的基因標誌 (gene markers)以及已知的連續型性狀外表型，建立統計模型，進而憑藉基因型推測出育種價估計值 (genomic estimated breeding values，GEBVs)，從中選拔出適合的自交系 (inbred lines)或育種計畫中的雜交組合 (hybrids)。 統計模型的建構中，如何只透過基因型資料，選擇適當的個體當作訓練集 (training set)進行外表型分析，建構出表現好的預測模型，在基因體選拔是個重要的議題。在本文的研究中，分析兩種方法：A-最適準則 (A-optimality)與D-最適準則 (D-optimality)兩種判斷方法，原理是試圖挑出最大變異的個體作為適合的訓練集。我們使用四組不同的作物基因資料，分別使用模擬結果與實際資料，並與之前研究的其他方法相比較，兩者相較於隨機訓練集有比較好的表現。

Genomic selection (GS) has become a powerful tool in the domains of plant and animal breeding with advanced and cheaper molecular genetic technology. Despite substantial reduction in genotyping costs, phenotyping still remains a time-consuming and expensive process. As a result, phenotype estimation through genotypic information can accelerate the breeding cycle. In GS, markers of the whole genome are used to estimate genomic estimated breeding values (GEBVs) by statistical models, which are built with genotype and phenotype. These GEBVs facilitate the selection of desirable inbred lines or hybrids for further breeding programs. In the construction of statistical models, selecting appropriate individuals as the training set based on genotype data and building effective prediction models is a crucial topic in genomic selection. In this study, we evaluated two methods: A-optimality and D-optimality, which are criteria aimed at selecting individuals with the highest level of variation. We utilized four different crop genomic datasets and compared the results with previous studies, using both simulated and real data. Both A-optimality and D-optimality demonstrated better performance compared to random training sets.