利用基因共同表現網路找出穩定的調控基因群

Abstract

為了瞭解基因調控機制，我們選擇微陣列生物晶片技術來偵測基因表現並且將得到的資料加以分析。傳統上，統計方法可以幫助我們選出在實驗資料中有差異表現的基因。然而，針對相同的研究主題，不同研究團隊的實驗資料常常會找到幾乎不相同的基因。這些結果再再顯示，此統計分析方法所找出的基因無法應用於不同資料組間，造成分析結果的穩定性不佳，無法幫助研究學者建構出通用的調控模型。 要解決統計方法的不穩定性，目前有關微陣列的資料分析已經逐步轉向基因群(gene set)的概念，而應用此想法的主要分析有兩大類：路徑分析(pathway analysis)以及網絡分析(network analysis)。路徑分析主要大量利用過去文獻紀錄以及預先定義好的與失調(dysregulation)相關基因群來解釋基因轉錄資料(transcriptomic data)。另一方法網絡分析則不需要利用過去知識，利用數學計算找出基因與基因間的連接關係，並藉此定義出與表現型相關的表現基因群。本研究的目標，在於找出穩定的調控基因群以及其內部未被詳加研究的基因；因此，我們將利用網絡分析方法來更進一步分析生物晶片的資料。 在本研究當中，我們利用權重型基因共同表現網絡分析(Weighted Gene Co-expression Network Analysis, WGCNA)來幫助我們建立非關尺度網路(scale-free network)，進而找出高度相關的基因群。在我們使用的肺腺癌資料中，網絡分析的確可以有效找出具有調控功能的基因群模組。這些被選中的模組，功能上可以分為幾大類：細胞骨架建構(cytoskeletal construction)、細胞週期(cell cycle)與免疫缺失(immunodeficiency)。而與免疫相關模組當中，有一個基因群之功能與B細胞受體調控(B-Cell Receptor signaling)高度相關，並且和肺癌病患存活也有高度相關。此B細胞受體調控相關模組的核心表現基因，被預測可能同時受到相同的轉錄因子Oct-1的調控。 總而言之，研究的結果證實，網絡分析的確可以幫助我們在不利用過往已知生物知識的情況下，有效建立出在不同資料間穩定的調控模型，而這些調控模組在生物上的確具有功能性。基於本研究證明，這樣穩定的分析可以提供生物研究學者另外找到可靠的研究基因的方法。

To understand the mechanisms of gene regulations, we usually use microarray technology to detect levels of gene expressions and analyze these microarray data. Traditionally, statistical approaches are implemented to select genes differentially expressed in the experimental data. However, for the same research topic, researchers discover almost different genes through different datasets. These results indicate no stability across datasets to construct a general regulatory model. To overcome this difficulty, currently the analysis has been turned into the gene set based solutions: pathway analysis and network analysis. Pathway analysis interprets transcriptomic data based on prior knowledge and pre-defined gene sets with dysregulations. On the other hand, network analysis identifies the gene-gene interactions without prior knowledge and searches for modules associated with phenotypes under research. In this study, we utilize the Weighted Gene Co-expression Network Analysis (WGCNA) to construct scale-free network for exploring highly correlated gene sets. From our lung adenocarcinoma training datasets, network analysis actually identifies regulatory modules. These selected modules could be categorized to be several functions: cytoskeletal construction, cell cycle regulation and immunodeficiency. Speaking of immune related modules, one module has been proved to be specific to lung cancer survival prediction, and it is annotated as B-cell receptor (BCR) signaling related module. The hub genes of BCR signaling related module has been predicted to be commonly regulated by transcription factor Oct-1. Based on our results and previous studies, we tried to propose a regulatory model concerning to lung carcinogenesis. In conclusion, these data indicate network analysis could help us construct stable regulatory network across datasets without prior knowledge, and the selected gene sets are biologically functional to suggest reliable research targets.