Please use this identifier to cite or link to this item:
標題: 利用混合分布模式之求配由兩自交系之雜交試驗估計數量性狀之基因效應
A Mixture-Modeling Approach to Estimating the Effect of Genes Affecting a Quantitative Trait from a Biparent Cross of Inbred Lines
作者: 張文怡
Chang, Wen-Yi
關鍵字: Mixture-Model
Quantitative Trait
Effect of Genes
出版社: 農藝學系
摘要: 本研究之結果顯示:在雜交試驗中,控制一數量性狀在兩親本間之差異的基因效應,可藉由取自兩親本(P1及P2)、雜交第二代(F2),及/或雜交第一代(F1)的數據,利用混合分布模式之求配來估算。此外,由混合分布模式之求配也可以估計有關基因座間的重組率。 我們以一項大麥雜交試驗所獲得之株高數據(Leonard et al., 1957)為例,說明利用混合分布模式之求配估計這些參數的過程。首先,利用SAS/STAT所提供的集群分析程序(FASTCLUS),將取自F2族群的觀測值分為幾個子群,再以各子群的參數估值做為求配混合分布模式的起始值,藉由限制的EM演算法尋找模式中各項參數的最大概度估值。其次,以McLachlan and Basford (1988)提出的方法計算近似訊識矩陣(approximate information matrix)的反矩陣,藉之估計各參數估值的變方變積矩陣,或藉由參數靴環自助法(parametric bootstrap method)獲得的99個由自助重抽樣品估得的估值,直接估算這些參數估值的變方變積矩陣。由此數例的結果顯示:這種估計程序比Powers之析出法(Powers, 1963)可以由數據中獲得更多的訊息。 我們也藉由模擬試驗探討各參數之點估值及信賴區間的統計特性。結果顯示,藉由混合分布模式之求配估計遺傳參數確實是一個可行的方法。當缺乏F1族群的試驗數據時,對估計的精密度影響不大。而在估算參數估值之變方變積矩陣時,則是以參數靴環自助法所求得的變方變積矩陣較佳。
In this study we show that the effects of the genes contributing to the difference in a quantitative trait between the two parents (P1 and P2)in a cross experiment can be estimated by fitting a mixture model by the data sampled from P1, P2, F2, and/or F1 populations. Additionally, this approach also gives estimates for recombination rates among the loci involved. As an example, we applied this estimation procedure to analyze the data of plant height in a biparent cross of barley (Leonard et al., 1957). The F2 data was firstly split into groups by a clustering procedure (FASTCLUS) provided in SAS/STAT. Taking the estimates given by the clustering analysis as starting values, a constrained EM algorithm was employed to find the maximum likelihood estimates for the parameters of the mixture model. The covariance matrix of the estimates was then obtained either by inverting the approximate information matrix (using the method developed by McLachlan and Basford, 1988), or by using a parametric bootstrap procedure with 99 replications. The numerical example seems to indicate that the present approach extracts more information than by using Powers' partitioning methods (Powers, 1963). A simulation was performed to investigate the statistical behavior of the point estimates and confidence interval for the parameters obtained by this approach. The result indicates that this procedure is feasible and useful. It also shows that no great loss in precision will occur if data of F1 are not available, and the covariance matrix obtained by bootstrap resampling is far better than that obtained from inverting the approximate information matrix.
Appears in Collections:農藝學系



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.