Information Scientific Foundations of Knowledge Discovery from Proteome Data

该研究为利用蛋白质组数据进行计算知识发现提供了信息科学基础，这些数据来源于全基因组蛋白质及其相关信息的分析。我们的贡献分为两个主题：1。推断蛋白质网络我们开发了一种统计和计算方法，通过使用微阵列基因表达数据和蛋白质-蛋白质相互作用数据来推断和估计基因网络。该方法采用贝叶斯网络和非参数回归相结合的方法，以Y2H和MS得到的蛋白-蛋白相互作用信息作为先验。我们还开发了一种计算方法，通过将主成分分析应用于微阵列基因表达数据来估计蛋白质复合物。此外，我们构建了一个基因调控网络的贝叶斯网络和蛋白质网络的马尔可夫网络相结合的模型。利用该模型，可以从蛋白质-蛋白质相互作用信息和微阵列基因表达数据中估计出更准确的蛋白质网络。路径建模与仿真的计算方法我们定义了一种新的基本体系结构，称为混合函数Petri网，并在此基础上扩展开发了一种新的路径建模与仿真软件工具。这种结构允许我们在通路建模中使用蛋白质亚细胞定位信息和蛋白质修饰信息。利用文献资料和随附数据，我们利用软件工具构建了包括蛋白质网络在内的通路模型。我们对这些模型进行了解释和评估，并通过对细胞凋亡信号通路、分裂酵母细胞周期通路模型和p53周围网络的建模证明了该方法的有效性。

This research developed information scientific foundations of computational knowledge discovery from proteome data which came out from the genome-wide analysis of proteins and their related information. Our contributions are classified in two subjects :1.Inferrining protein networksWe developed a statistical and computational method for inferring and estimating gene networks by using microarray gene expression data and protein-protein interaction data. This method uses a Bayesian network method combined with nonparametric regression which employs protein-protein interaction information obtained by Y2H and MS as its prior. We also developed a computational method for estimating protein complexes by applying the principal component analysis to microarray gene expression data. Further, we constructed a model which combines Bayesian networks for gene regulatory networks and Markov networks for protein networks. With this model, more accurate protein networks can be estimated from protein-protein interaction information and microarray gene expression data.2.Computational method for pathway modeling and simulationWe defined a new basic architecture called Hybrid Functional Petri Net with extension on which a new software tool for pathway modeling and simulation was developed. This architecture allows us to use protein subcellular localization information and protein modification information in pathway modeling. By using the literature and attached data, we constructed models of pathways including protein networks with the software tool. We interpreted and evaluated these models and demonstrated the effectiveness of this method through modeling of a apoptosis signaling pathway, a cell cycle pathway model of fission yeast, and a network around p53.