訓練集最佳化用於建立加性加顯性效應預測模型

Abstract

隨著世界人口的增加，糧食安全成為大家關注的焦點。其中農業作物的生產在糧食安全中扮演重要的角色。而在作物育種上，最常使用增加產量的育種方法為雜交育種，透過利用雜種優勢來增進產量。然而雜交育種隨著評估的親本數增加，所產生潛在的優良雜交種數目就會大幅增加，也使得過程十分耗費人力與時間。基因體選拔的出現能夠使得雜交育種所需的成本大幅降低，利用個體的基因資訊建立基因組預測模型來評估雜交後代表現或親本的評估。而其中如何選擇建立基因組預測模型的訓練集就十分重要，但針對雜交育種的加性加顯性效應模型下最佳化訓練集的方法較少，因此本研究希望能夠透過將現有的方法拓展到加性加顯性效應模型，並能有效的增進預測效率。研究透過分析玉米、小麥與水稻資料集來評估，結果顯示三個基於全基因體迴歸模型的方法 (PEV、r-score 與 MSPE-score) 能夠在雜交育種作為一個有效的訓練集最佳化方法。

Food security has become a major concern as the world's population increases. The production of crops plays an important role in food security. In crop breeding, the most commonly used breeding strategy to increase yield is hybrid breeding, which takes advantage of heterosis to increase yield. However, as the number of parents evaluated increases, the number of potentially superior hybrids increases dramatically, making the process very labor-intensive and time-consuming. The advent of genomic selection has made it possible to significantly reduce the cost of hybrid breeding by using individual genetic information to build genomic prediction models to evaluate the performance of crosses or parental lines. However, there are few methods to optimize the training set for the additive plus dominant effects model of hybrid breeding. Therefore, this study aims to extend the existing methods to the additive plus dominant effects model. The study was evaluated by analyzing maize, wheat, and rice datasets, and the results showed that three methods (PEV, r-score, and MSPE-score) based on the whole genome regression model could be an effective training set optimization method in hybrid breeding.