水稻基因組選種之模擬研究 ── 訓練族群預測模型之建立與最低投入試驗規模之確立

Abstract

基因組選種 (genomic selection) 是新興的分子標記輔助選種策略，透過由訓練族群所建構的統計預測模型，直接以個體大量的分子標記基因型資料計算各個個體的個體育種價估計值，並以此為依據選拔個體。已知預測模型的統計方法、分子標記數量以及訓練族群大小，皆會影響個體育種價估計值的預測準確度。本研究挑選計算能力優異的RR-BLUP、BL與RKHS三種統計方法建構預測模型，並依據有效基因座數目、訓練族群大小、分子標記數量、和性狀狹義遺傳率等四種參數的不同設定，模擬192種不同水稻重組自交系訓練族群的基因型與外表型的資料。然後計算192種組合模擬資料之三種統計模型的預測準確度，評估與比較各種參數設定對預測準確度的影響，以決定投入訓練族群之試驗規模。評選方法是依據不同的狹義遺傳率，先選取基因組選種之預測準確度可高於外表型選種的所有組合，再由這些組合中選出最小訓練族群大小且最少分子標記數量的組合。

Genomic Selection is a new strategy of marker-assisted selection that selects superior individuals based on their genomic estimated breeding values. The genomic estimated breeding values are calculated solely using individual genotypes of substantial markers through a statistical prediction model built by data collecting from a training population. Prediction accuracy of genomic estimated breeding values can be affected by several factors, including statistical methods of the prediction model, number of markers genotyped, and size of the training population. In the current study, three statistical methods – RR-BLUP, BL, and RKHS – all of which have great computing ability were chosen to establish the prediction model. 192 different sets of genotypic and phenotypic data of rice recombinant inbred populations were simulated in silico as training populations among which effective QTL numbers, population size, marker numbers, and narrow-sense heritability were assigned at different levels. In order to determine the most effective inputs of a training population for given narrow-sense heritability of a characteristics, prediction accuracy of genomic estimated breeding values was calculated and compared for all simulated training populations using the three statistical methods. At each different level of narrow-sense heritability, sets of training populations showing that genomic selection is more effective than phenotypic selection were identified, and then the set with lowest marker numbers and smallest size of the training population were selected.