基因富集檢定方法之比較

Abstract

微陣列可同時偵測數以萬計的基因表現量，要如何在繁多的基因中挑選出重要且具代表性的基因為大家致力的目標。挑選具代表性基因的研究單位為單一基因，著重於檢定每個基因在一個微陣列資料中不同條件下的表現量是否顯著不同，再將較顯著的基因挑選出。相較於關注研究單一基因的表現，而後衍生出以探討一群基因顯著性為目的的研究，即為基因富集檢定 (gene enrichment test)。基因富集檢定研究對象為一群根據研究目的定義而選取的基因稱為基因組，目標是分析選定基因組內的基因在不同條件下表現量是否顯著不同。基因富集檢定已發展出許多統計方法，常用的基因富級檢定方法大致可分為兩類：單變量方法與多變量方法。單變量方法彙整基因組內各基因單變量檢定統計量的結果作為基因組顯著與否的依據，而多變量方法視基因組表現量資料為高維度的常態分佈進行統計分析，得以考慮基因間的相關性。論文中先以模擬資料比較常用的基因富級檢定方法，以檢定力 (power) 與曲線下面積 (AUC) 做為優劣比較的指標，並以實際資料比對不同方法富集檢定的結果。結果顯示霍特林T方及基因集合富集分析在模擬資料及實際資料皆能檢定出最多顯著基因組，而魏可遜等級和檢定及柯史檢定的敏感度及特異度最高，綜合來說，基因集合富集分析穩定且準確度高，為表現最佳的方法。

Microarray aims to simultaneously monitor the expression of thousands of genes. It is usually the objective to mine important information from the data, such as the representative genes that differentially expressed (DE) under different conditions. In recent years, several gene set enrichment tests have been proposed to search for a DE gene set under different conditions. The gene set enrichment tests can be divided into two categories, univariate and multivariate methods. The former summarizes univariate statistics from each gene in the set to infer whether the gene set is significantly DE or not, while the latter considers the correlation among genes by assuming multinormal distribution and using multivariate analysis. In this study, we compared seven gene set enrichment tests by simulations. The tests were also practiced on a real microarray dataset. The results showed that Hotellin's T2 and gene set enrichment analysis were the most powerful. Wilcoxon rank sum test and Kolmogorov-Smirnov test had the best sensitivity and specificity. In conclusion, gene set enrichment analysis was the most robust method to detect DE gene sets.