利用變異數組成模式對數量性狀基因座的貝氏統計推論

目標：利用貝氏的點估計值來提供一個估計主效基因的變異數的更簡便方法；一旦估計值可被確定，所利用之標識基因可被判斷是否與主效基因有連鎖，再進一步作基因定位分析。方法：將數量性狀的變異數分成主效基因的變異、主效基因以外的其他遺傳變異和環境的變異等三個成份來討論，再利用貝氏方法及最大概似法對三個變異數成份作估計以進行統計推論。事後樣本的取得會利用Gibbs抽樣方法以及Markov chain Monte Carlo法。結果：以核心家庭有兩位子女的情況作模擬，經由結果顯示，對於主效基因的變異數估計，貝氏方法確實比最大概似估計法較穩定且精確。結論：相較於最大概似法，貝氏分析確實提供了精確的估計方法，而且本文利用WinBUGS1.3的程式可以獲得事後機率的樣本，可利用這些事後樣本做進一步的包含機率意義的統計推論，這是傳統統計方法所無法達到的目標。若是推廣到較複雜的遺傳模式，如主效基因不只有一個，或者改變事前分配，則此遺傳模式也可以很容易地被改寫，再利用現有統計軟體計算出結果，作為未來進一步定位之用。

關鍵字

貝氏統計推論；主效基因；數量性狀；變異數組成模式； Gibbs抽樣方法

並列摘要

Objectives: To estimate more efficiently the variance of the major gene via Bayesian method. Methods: The variance component model is considered for the quantitative trait. The model comprises a polymorphic single major gene, polygene, and a random environmental effect. Two different approaches, maximum likelihood and Bayesian estimate, are compared in terms of the estimation of the three variance parameters of interest. The inference of the posterior distribution and posterior samples of the parameters, particularly the additive variance of the single major gene, are made via Markov chain Monte Carlo method using WinBUGS1.3. Results: Simulations are conducted to compare the performance of MLE and Bayesian estimates. Conclusions: The results show that the Bayesian point estimate, the posterior mode, is more accurate than the MLE. The posterior variance is also smaller than that of MLE. Compared with the conventional maximum likelihood estimation, the Bayesian approach is more flexible when a more complex genetic model is considered.

並列關鍵字

Bayesian ； major gene ； quantitative trait ； variance component model ； Gibbs sampling

參考文獻

(2000).Practice Management Information Corporation (PMIC).International Statistical Classification of Diseases, 9th Revision, Clinical Modification, Sixth Edition.