EM-AMMI應用於整合多年期作物區域試驗分析

Abstract

區域試驗施行於品系產量試驗之後，其目的為確保品系之產量與農藝性狀在不同環境下皆能具有穩定表現。當區域試驗資料的基因與環境交感效應顯著時，可採用AMMI模式對基因與環境交感項做進一步之分析，能提升品系選拔的效率。惟使用AMMI模式分析區域試驗資料時，不允許基因型與環境組合的平均產量有缺值的情形。然而AMMI模式奇異值分解的不允許基因型與環境組合的平均產量有缺值的情形，但多年期多重地區之區域試驗資料通常高度不均衡，限制育種家探討跨年期間基因型與環境的交感。本研究利用EM-AMMI方式進行缺值估計。模擬結果顯示，無論缺值率大小，僅使用第一主成分軸之缺值估計效果為最佳。本研究亦應用EM-AMMI於跨年期毛豆區域試驗資料之品種穩定性比較，希望藉此幫助育種家進行較長期的育種材料評估。 The regional trials aim to ensure the stability of yields and agronomic traits of the targeted lines under various environments. When there exists a significant genotype-by-environment interaction, AMMI model can be applied to improve the efficiency of selection. However, AMMI model is not applicable when the average yields are missing in some genotype and environment combinations, which cannot be avoided when combining multiple regional trial data to study the genotype-by-environment interaction across years. In this study, we adopted Expectation-Maximization-Additive Main Effect and Multiplicative Interaction (EM-AMMI) method to impute the missing average yields. According the simulations, imputation using only the first principle component axis had the most accurate estimates regardless of the missing percentages. We also applied EM-AMMI method to the multiple-season vegetable soybean regional trial dataset and then proceed to AMMI analysis on the imputed dataset. The proposed method may make it possible for the breeders to perform long-term evaluations on their breeding materials.