校準模式在Affymetrix基因晶片資料上的應用

Abstract

生物晶片已成為一項被生物學家廣為使用的工具，其特點是能一次研究大量的基因或蛋白質的表現。晶片實驗流程複雜，造成晶片資料存有許多系統誤差及未知的干擾因子，再加上成本昂貴，不易增加實驗重複數;因此如何正確有效地由生物晶片資料中獲取資訊及採用適當的重複數，這一直是眾人所關心的問題。 Affymetrix 公司所生產 Genechip(oligonucleotide array) 是目前市面上精準度及重現性(reproducible) 較高的晶片。因為特殊的晶片設計，原始資料須先經過背景值校正、正規化、表現量轉換等步驟獲得基因表現量值，之後才進行近一步分析。目前已有提出許多獲得表現量值的方法，常被使用的有MAS5.0、MBEI、RMA等方法。本研究提出一套以校準模式(calibration model) 為基礎之方法，可用來適當分析 Affymetrix 晶片資料。藉由分析一組實際資料，結果顯示校準模式跟其他方法相比具有相當的正確性。 以校準模式為基礎，修改 Hess and Iyer(2004) 的模擬方法，在給定少量實驗晶片下，藉由適當的參數設定產生模擬實驗資料。模擬資料除了可比較不同表現量轉換及分析方法，也以此為基礎，提出一套流程來適當選取晶片。

Affymetrix high-density oligonucleotide biochips have been widely used in various biological and medical experiments. One of the important processes in analysis of such experimental data is to summarize the probe intensities corresponding to a unigene into a single gene expression value. The gene expression value obtained from different summarization methods can lead to controversy results in identification of differentially expressed genes, and in further analyses such as clustering analysis and gene network analysis. In this study, we propose a new summarization method to obtain gene expression values based on a modified calibration model presented in Rocke and Durbin (2001), which is originally used to analyze the data of gene level. We adapt it for the data of probe level. The proposed method could compete with some existing methods (RMA, MBEI, MAS5.0) according to the results of analyzing a spike-in data set provided by Affymetrix. Furthermore, we develop a simulation mechanism to mimic naturally occurring data based on the modified calibration model. Finally, we offer a method of determining the sample size for the test-control experiment using Affymetrix GeneChips.