以排列法篩選重複測量微陣列晶片資料中的顯著基因

Abstract

重複測量試驗設計在研究基因調控路徑上有很多好處，重複觀察多個基因在不同時間點上的表現，將可獲得各個基因表現的先後次序之資訊，而發生時間的先後為建構基因的因果關係之基本要件。最近十年來，基因微陣列技術也對生物相關領域的發展有莫大的幫助；然而，目前基因微陣列實驗的成本仍然很高，大部份的重複測量基因微陣列實驗僅有幾個生物體的重複。由於許多重複測量的分析工具都是基於大樣本理論的架構下發展出來的，這些方法在小樣本資料的應用上通常表現不佳，因此也就不適合應用於重複測量基因微陣列資料的分析上，包括近年來廣泛使用於分析相依資料的廣義估計方程式(GEE)方法。 我們提出使用GEE合併「排列法」來處理GEE在小樣本資料表現不佳的問題。電腦模擬的結果顯示，「GEE合併單變數排列法」並使用以模式為基礎的變方估計式(Model-based variance estimator)，在控制名義上所宣告的第一型統計錯誤及維持相對高的統計檢定力上有很好的表現。假如樣本數十分少，例如：少於五個時；我們則建議使用「GEE合併多變數排列法」並使用以模式為基礎的變方估計式，進行篩選重複測量微陣列資料上的顯著基因，這樣的分析架構使得在控制一定數目的偽陽性(False positive)下，可維持相對高的偵測顯著基因之能力。

Repeated measurement design has lots of advantages on the investigation of underlying genetic pathway. Recently decade, microarray technology also has great aid of improvements in biology relative fields. Because the cost of microarray is still high, most of microarray experiments with repeated measurement design are only several biology replicates. Many repeated measurement analysis tools are based on asymptotic theory, the small samples performance of these methods are often unsuitable to microarray repeated measurement data including the popular generalized estimating equations (GEE) method for analysis of correlated data. We suggest by using GEE combining with permutation methods to solve the problem. The simulation results show that model-based variance estimator with univariate permutation GEE to analyze repeated measurement microarray data performs well on the controlling of nominal type I error with maintaining relative high power. If the sample sizes are extremely small, e.g., less than 5, we propose to use model-based variance estimator with multivariate permutation methods to control the number of false positive with maintaining relative high detective ability.