具穩健性之雙染色cDNA微陣列試驗設計之研究

Abstract

微陣列晶片是近十年來興起並廣泛應用的生物技術。相較於傳統的序列表現量檢定技術，微陣列晶片能夠同時檢測上萬個mRNA片段的表現量，大幅提昇了實驗的效率。在應用上，為生物學家找尋功能基因(functional genes)及基因調控網路(gene regulatory network)提供強而有力的協助。相對於微陣列晶片的檢測能力，其高維度的基因表現資料(gene expression data)既龐大且複雜。在微陣列試驗過程中，觀測值發生缺失是難以避免的狀況。對於缺失的資料，有兩種常見的處理方式：(1)對缺值進行估計，再以全體資料進行分析。(2)刪去缺值，利用剩下的數據進行分析。目前已經有許多研究探討如何在資料分析前進行缺值估計(Troyanskaya et al., 2001; Nguyen et al., 2004)。而本篇研究主要是利用試驗設計的方法，針對後者作處理，也就是刪去缺值的處理方式進行討論。 在雙染色cDNA微陣列實驗的系統下，本研究考量了忽略染劑效應以及將染劑效應納入變異來源的兩種不同模型，各模型分別再以比較實驗(comparative experiments)及處理對照實驗(test-control experiments)這兩種常見的實驗類型逐一探討。 一個適當的試驗設計可以協助實驗者在有限的資源下進行實驗，並獲得最大的資訊解答欲知的問題。在有資料缺失的限制下，刪去發生缺值的晶片資料後，一個不完全區集設計仍維持所有對比(contrast)的可估性(estimable)，即稱為一個符合穩健性第一準則(robustness criterion I)的設計(Goash, 1979)。本研究中，將穩健性第一準則的概念進行延伸，定義缺失型態(flawness pattern)、喪失連接性之最小缺失晶片數(breakdown number)以及維持連接性之剩餘設計的比例(proportion of connected residual designs)作為一設計穩健程度的判定指標。並利用以上三種指標，比較不同設計之間的穩健程度高低，最後將判定結果及建議設計列於本文附表，提供微陣列實驗者參考利用。

Microarray has been an increasingly popular biotechnology in measuring gene expression of a biological sample. In contrast to the traditional methods, it enables biologists to evaluate ten thousands of mRNA sequences in expression simultaneously. Statistical methods developed for such large-scale and complex data from microarray experiments have been intensively explored within recent years. On the other hand, the studies on the design issues need further investigation. In the current study, we focus on one of the important design issues in two-color mciroarray experiments. We propose a new criterion to evaluate the robustness of designs against missing data. Furthermore, we construct the robust designs suitable for the two-color microarray experiments. Missing data are frequently confronted during a microarray experiment. Conventionally, one performs analysis either estimating or ignoring missing data. However, we are currently interested in selection of good designs that may provide high robustness against missing data. We first present two linear models in characterizing the data of a two-color microarray experiment according to whether or not take into account the variation between two fluorescent dyes. Then we seek for the robust designs based on the proposed models and the robustness criterion. The criterion we proposed is to compute the proportions of the connected residual designs for the class of block designs of size two and that of row-column designs with two rows. The robust designs are defined as those who have the maximal proportions if a number of blocks (columns) are missing. In specificity, we investigate two kinds of experiments in this thesis, including treatment comparative experiments and test-control experiments. Two classes of practical designs are respectively proposed for these two kinds of experiments using two-color microarrays.