利用已知基因傳遞機制及蛋白質交互作用圖譜來開發新的基因互動途徑

Abstract

近年來，基因晶片 (microarray) 的數據分析已經從傳統上僅利用統計分析的方法，轉向加入更多已知基因功能性註解來協助分析。生物反應路徑分析法 (pathway analysis) 針對資料庫中每一個已定義好的反應路徑 (pathway) ，測量於實驗設計中其是否存在訊息核糖核酸 (mRNA) 層級上的顯著變異。而另外一種網絡圖譜分析法 (network analysis) 則旨在搜尋一個由基因間所有可能存在的交互作用所組成的圖譜 (global interaction network) ，看其中是否有顯著改變的子圖譜 (subnetwork) 。這兩種分析方法各擅勝場，且可望補足對方的缺點：前者僅分析已知的反應路徑，故將結果局限在熟知的生物知識中；後者雖蘊藏了許多可能的基因互動途徑，但直接從 global interaction network 中搜尋容易找到無法以現有生物知識呈現其一致性生物意義的subnetwork。 本文提出一個能基於pathway analysis但更進一步結合network analysis優點的分析方式來改善現有的分析方式，值得一提的是此一結合概念在目前的分析領域中並不常見。 此方法首先利用 Tian et al. 發展的pathway analysis方式測定有顯著變異的pathway。接著以這些pathway的成員作為出發點，採用Nacu et al. 的network analysis方法進行一個目的導向式的subnetwork搜尋。 本篇論文將此方法應用在台大醫院肺癌病人的基因晶片數據上。一開始嘗試在有變異的pathway中尋找其最具代表性的成分 (subnetwork) 。這組數據產生的subnetwork在另一組台北榮總肺癌病人的數據中亦得到了呈現高度一致的結果。此外，我們針對找到的subnetwork進行其成員基因的功能性分析，發現從原本完整pathway縮減到subnetwork的過程中，整體的功能由原本具有的多樣性，專一化到特定的功能上。這暗示了該pathway的某部分功能在實驗中是明顯地被改變的，而這樣的改變得以用這篇論文的方法被察覺。更進一步，我們展現了本方法承繼了network analysis而來的優點。立基於pathway的已知成員去搜尋其可能有互動的鄰近基因，我們得到的subnetwork是以此pathway為出發點，但包含許多未被認識的交互作用，這樣的結果可以協助研究者對於未知的部分設計實驗做更深入的探究。從另外一方面來說，這樣的結果也有異於傳統network analysis的方式，它能使得研究可以立基於研究者感興趣的pathway，基於已知的生物知識去拓展相關未知的可能性。 總結所有分析的結果，它們從不同方向指出了此分析方法不同於以往的許多優點：它可以從顯著改變的pathway中萃取出一個最重要且大小適合研究的subnetwork、也可以針對研究者感興趣的pathway或特定的調控機制進行主題式的深入探討、此外除了立基於原有生物知識外，它亦具有開發基因間新的互動機制的能力。

Currently the analysis of microarray data had turned into integrating with prior biological knowledge: pathway analysis interprets transcriptomic data on pathway level and identified predefined groups of genes with dysregulation; network analysis takes gene-gene interactions information into consideration and searches for modules associated to the phenotypes under study. The two analyses have its own advantages respectively and they complement the weaknesses of each other: pathway analysis provides little clues to directly explore new biological knowledge and network analysis usually yields modules including few consistent biological information. In this study an analytical methodology was developed to integrate current pathway analysis method with network analysis methods. Initially, dysregulated pathways are identified by modified pathway analysis method in Tian et al.. Subsequently, a focus-oriented investigation on dysregulated pathways are performed by network analysis following the work of Nacu et al., and this step is using modules within or related to members of the pathways to be further investigated. Several improvements were made, such as the scoring functions and the module identification algorithms. To illustrate the benefits of this methodology, a lung cancer study with 30 paired cancer and normal tissues was explored. The results derived within dysregulated pathways were also identified consistently in another public dataset GSE7670. Furthermore, GO term enrichment analysis was applied to show that the modules have a specialized functionality than the original pathways. In brief, original large modules were reduced from the entire pathway to a smaller size of relevant interconnected members, which are much easier to be manipulated but still remain their biological information. Moreover, the ability of this methodology to explore novel interactions related to pathway members were also demonstrated by extending the module search algorithm beyond the pre-defined pathways. This would not be achieved by traditional pathway analysis methods, which usually don’t include biomolecular interaction information. Yet, modules identified in this methodology were based on dysregulated pathways with specific biological meaning since their members were mainly associated. In conclusion, these data all indicated the advantages to integrate both pathway and network information during microarray analysis: to uncover manageable size of molecular interaction networks important for pathway dysregulation, to focus on interested pathways, functions or even specific regulatory events, and to possess the potential of performing exploratory researches on mechanisms that are not yet well understood. Undoubtedly, this concept could be extensively applied to other array experiments of similar design regardless of the disease under study.