從候選族群中選拔最佳基因型

Abstract

從大量種原庫中選拔出優良種原是作物育種計劃中非常重要的一環，而基因組預測(genomic prediction)在此是一項非常有效的工具。本研究提出了一種迭代策略，使用基因組預測的方式從大量候選種原庫中選拔最佳基因型，而候選種原的個體在未知外表型的情況下進行基因組預測。首先從候選種原中隨機抽樣少部分的外表型測量(phenotyped)，且使用數據建立基因組BLUP(GBLUP)預測模型，採用預期增進(expected improvement, EI)準則來判斷是否已選拔出最佳基因型，如果在當前步驟未達到目標，則從剩餘子集合中的未知外表型個體來選拔且進行外表型調查，然後將外表型數據與當前訓練數據一起加入且更新GBLUP預測模型，重複該過程直至選拔出最佳基因型。我們透過真實數據集和模擬數據集，比較基於育種價(breeding value)或基因值(genomic values)分佈的EI準則下的迭代策略結果，我們發現相較於育種價，基於基因值的EI準則只需要較少的基因型(訓練集大小)即可找到最佳基因型。本研究使用了水稻，玉米，小麥和南瓜的五個真實基因組數據集。

Genomic prediction has been a powerful tool to select superior accessions from a large germplasm collection in a plant breeding program. An iterative strategy using genomic prediction to discover the best genotype from a large candidate population is proposed in this study. The individuals of the candidate population have been genotyped only without being phenotyped yet. A genomic BLUP (GBLUP) prediction model is first built using the phenotype data for some individuals randomly sampled from the candidate population. The expected improvement (EI) criterion is employed to determine whether the best genotype is discovered. If the goal is not achieved at the current step, then a suitable subset from the remaining non-phenotyped individuals is chosen and phenotyped. The phenotype data are then added with the current training data to update the GBLUP prediction model. The procedure is repeated until the best genotype is discovered. The EI criterion based on the distribution of breeding values or genomic values is investigated for the iterative strategy, through several real datasets and simulated datasets. Our proposed EI criterion based on the genomic values is shown to frequently have less genotypes required (the training set size) to find the best genotype than that based on the breeding values. Five real genome datasets of rice, maize, wheat and pumpkin are used in this study.