微陣列基因表現資料一致性的統計方法之評估研究

Abstract

微陣列是在21世紀裡突破性的技術之一。但不管它有多大的潛力，以微陣列為基礎所發展出來有用於臨床實驗的產品發展的速度卻不迅速。直到最近，美國食品和藥物管理局才批准這種基於微陣列技術的第一個生物晶片產品。主要原因之一在於從微陣列實驗所獲得的基因表現量缺乏一致性和再現性。有鑑於此，一致性和再現性最近已經在微陣列實驗過程中引起許多研究探討。一般來說習慣使用皮爾遜相關係數來評估微陣列表現量在不同重複之間的一致性。然而，皮爾遜相關係數是用來評估兩個變數之間的相關性，但並不適合去評估當同一組基因來自不同重複間的表現量程度的一致性。因此，我們建議使用一致性相關係數去評估在重複之間同一組基因表現量的一致性。我們也使用廣義樞紐量的觀念導衍一致性相關係數的精確信賴區間。另外，一致性相關係數的單尾信賴下界被用來檢定假設是否大於一個最小的可接受量。我們亦將進行模擬研究，在不同的參數平均差、變異數和樣本數的組合下計算覆蓋率、期望長度和檢定力，比較皮爾遜相關係數方法，一致性相關係數大樣本方法，及其精確信賴區間在評估一致性的優劣。最後提出一個實例數據加以說明。

Microarray is one of the breakthrough technologies in the twenty first century. Despite of its great potentials, the transition and realization of microarray technology into the clinically useful commercial products has not been as rapid as the technology could promise. Only recently, the US Food and Drug Administration (FDA) approved the first biochip product based on the microarray technology. One of the primary reasons is lack of agreement and poor reproducibility of the intensity measurements on gene expression obtained from microarray experiments. As a result, agreement and reproducibility have recently drawn a lot of attention in microarray experiments. Current practices often use the Pearson’s correlation coefficient to assess the agreement of expression data between the technical replicates from microarray experiments. However, Pearson’s correlation coefficient is to evaluate association between two variables and is not appropriate for evaluation of agreement of expression levels of the same gene between technical replicates. Therefore, we propose to use the concordance correlation coefficient (CCC) to assess agreement of expression levels of the same gene between technical replicates. We also apply the Generalized Pivotal Quantities (GPQ) to obtain the exact confidence interval for concordance correlation coefficient. In addition, a one-side (1-α) lower confidence limit for ρc is employed to test the hypothesis that agreement of expression levels of the same gene between two laboratories exceeds some minimal requirement of agreement. We conducted a simulation study, under various combinations of mean differences, variability, and sample size, to empirically compare the performance of different methods for assessment of agreement in terms of coverage probability, expected length, size, and power. Numeric data from published papers illustrate the applications of the proposed methods.