在本論文中，我們說明如何應用圖形理論對蛋白質交互作用網絡進行資料探勘，並藉此偵測必要性蛋白質與蛋白質功能模組。對於偵測必要性蛋白質的研究，我們利用必要性蛋白質與非必要性蛋白質各自的鄰居所形成之子圖間有明顯差異的特性，發展出三種偵測必要性蛋白質的方法。除此之外，我們提出一種綜合兩種偵測方法的技巧，以得到更好的預測結果。對於偵測蛋白質功能模組的研究，根據前人研究顯示，蛋白質複合體是由核心成員和附屬成員所構成，依據該結果，我們設計出一種分群演算法，利用找出蛋白質交互作用網絡中的聚落以預測蛋白質功能模組，該方法不僅可以處理帶有權重的網路，並可利用基因表現的資料，以得到更好的預測結果。除此之外，我們經由設計一個評估分群結果的指標，藉由該指標對分群結果汰弱擇強，以提供更好的分群結果，並將其運用在蛋白質功能模組的偵測研究上。There are many bioinformatic methods for predicting protein’s functions. In this dissertation, we show how to apply graph theory to a protein-protein interaction network to predict essential proteins and functional modules. Based on the neighborhood of an essential protein is usually larger and denser than that of a non-essential protein, we proposal three methods to predict essential proteins. We also design a double screening scheme, which combines the results computed by two different methods, to generate a superior result. For predicting functional modules, we develop a clustering method which not only extract functional modules from a weighted PPI network, but also use gene expression data as optional input to increase the quality of outcomes. We also propose a measure to judge a cluster and use this measure to develop a framework that integrates the different clustering results to produce a better result.
