生物复杂网络中功能模块的挖掘

发布时间：2018-05-10 23:19

本文选题：复杂网络 + 生物网络　；参考：《上海交通大学》2015年博士论文

【摘要】：网络科学作为一门交叉学科,它的基本理论正渗透到从数理科学到生命科学、工程科学甚至社会科学等众多学科中去。复杂网络的研究引起了世界不同领域科学家的广泛关注。对复杂网络的定性和定量特征的认识和理解是网络时代中一个重要而又具有挑战性的课题。作为复杂网络中一个重要特性,模块结构(或称社团结构)是一个重要而又普遍存在的结构特性。准确挖掘和分析模块结构对理解复杂网络的演化、结构和动态性都有着理论和实践的意义。模块结构作为生物复杂网络中的功能模块组织形式,在生命科学领域中有着重要的意义。人们虽然提出很多有效的算法来分析功能模块,如基于图论的方法,基于随机游走模型和谱聚类方法,但是这些方法在算法层面和生物网络局限性上都存在一定的缺陷。面对这些问题,需要我们有针对性地研究并提出新的功能模块挖掘方法。在本文中,我们主要研究了如何挖掘生物复杂网络中的功能模块,并探索了多样性的功能模块组织形式。首先,针对现有模块分析算法存在的缺陷,提出一种衡量网络中任意两个结点的新相似性ISIM,依据这种新的相似性和层次聚类思想对生物网络中的功能模块进行挖掘,并利用新结点相似性进一步揭示了生物网络中蛋白质复合物的层次性和功能模块的多尺度性。为了避免生物网络不完备性带来的影响,我们通过融合多条件下基因共表达谱数据构建完备的基因共表达网络,进而分析功能模块。接下来,我们突破模块结构是生物复杂网络中功能单元的唯一组织形式这一概念,在生物网络上发现一种新的功能模块组织形式,Bi-sparse功能模块。并相应地提出一种能同时挖掘高聚合和稀疏功能模块的二叉树搜索方法。最后,通过大规模网络的统计分析,深入研究了Bi-sparse模块的一些特性。因此,本文的研究内容和创新点主要包括以下几个方面:(1)为了有效地克服现有功能模块挖掘方法的缺点,我们使用受限的随机游走模型,提出一种新的转移概率矩阵,进而定义一种新的结点相似性ISIM来衡量网络中任意两个结点之间的距离。新的结点相似性有三个良好的特性,一是它能成功地融合网络的全局和局部拓扑信息;二是新结点相似性不仅能有效地衡量两个结点之间的距离,而且能捕捉到两个结点在网络中的拓扑结构;三是它是在一个收敛的空间定义结点相似性,因此,在一系列不完备和含有噪声的生物网络中,具有良好的稳定性和鲁棒性。使用新结点相似性和层次聚类思想,可以有效地分析生物复杂网络中的功能模块。首先,我们使用新结点相似性产生网络的相似性矩阵。然后,使用层次聚类思想建立网络模块的树状结构。最后,选择合理的目标函数自动地挖掘网络中的功能模块。在此基础上,通过改变新结点相似性中的调节因子,本文又提出一种新的方法(isimb方法)揭示生物网络中蛋白质复合物的层次结构和功能模块的多尺度特性。与现有的模块挖掘方法相比,基于新结点相似性的方法是一个无参数的方法,它能自动地确定网络中模块的个数。使用它挖掘到的模块不仅与真实的功能模块结构获得更好的匹配,而且能有效克服生物网络不完备性的缺陷。与单尺度的方法相比,本文创新性地把模块多尺度概念引入到生物网络中蛋白质复合物和功能模块的挖掘,这种新的理念不仅能成功地预测蛋白质复合物及其层次特性,而且能从具体到一般的视角揭示功能模块的动态过程。(2)针对生物网络的不完备特性和基因共表达的不传递性,本文提出一种新的方法检测基因共表达网络中的功能模块。这种方法首先融合不同条件下基因共表达谱数据构建完备的基因共表达网络,随后使用最大团算法挖掘网络中的功能模块。这种新的方法与其它方法相比,预测的结果有较强的生物功能相似性。通过转录和调控关系分析,预测功能模块中的基因有较高的概率被同一个转录因子所调控,从而为构建基因调控网络提供丰富的结果。(3)传统的生物网络中功能模块的挖掘都是基于高聚合的模块结构是功能模块组织的唯一形式。然而,这个结论在生物网络中,特别是在蛋白质相互作用网络中存在可疑性。因此,我们发现一种与高聚合模块不同的bi-sparse模块,然后结合二叉树理论和矩阵论提出一种新的方法(bts方法)来挖掘两种类型的功能模块。bts方法在蛋白质相互作用网络中挖掘的高聚合模块和bi-sparse模块都组成功能单元。与其它的方法相比,bts方法具有良好的性能:一是不需要预先设置模块的个数;二是挖掘的高聚合模块和Bi-sparse模块都具有显著性的生物功能相似性。(4)我们把高聚合模块和Bi-sparse模块作为功能单元的组织形式共存于同一网络中,这一概念进行泛化。我们整理了四种类型共25个网络,用BTS方法对25个网络中的模块进行分析,结果发现:(a)Bi-sparse模块具有普遍性;(b)在社会网络中,Bi-sparse模块中的人们充当着经纪人的角色,负责协调不同群体之间的矛盾,促进信息、技术和知识的交流等作用;在计算机软件网络中,Bi-sparse模块中的结点具有相似的软件包属性;在生物复杂网络中,Bi-sparse模块中的蛋白质或基因具有显著性的功能相似性;(c)复杂网络中的Bi-sparse模块拥有一些特性:一是Bi-sparse模块和高聚合模块相比,Bi-sparse模块含有的结点较少;二是Bi-sparse模块在不同类型的网络中,具有一定的偏好性;最后,存在复杂网络中的Bi-sparse模块有两种明显的拓扑结构。
[Abstract]:As a cross discipline, the basic theory of network science is permeating from mathematical science to life science, engineering science and even social science. The research of complex networks has aroused wide attention of scientists in different fields of the world. The understanding and understanding of the qualitative and quantitative characteristics of complex networks is the network age. An important and challenging task. As an important characteristic of complex networks, modular structure (or community structure) is an important and universal structural feature. It is of theoretical and practical significance to understand the evolution, structure and dynamics of complex networks. The organizational form of functional modules in complex biological networks is of great significance in the field of life science. Although many effective algorithms are proposed to analyze functional modules, such as the method based on graph theory, based on random walk model and spectral clustering method, these methods have some limitations on both the algorithm level and the biological network. In the face of these problems, we need to study and propose new functional module mining methods. In this paper, we mainly study how to excavate functional modules in the complex biological network and explore the functional modules of diversity. Firstly, we propose a kind of defects in the existing module analysis algorithm. The new similarity ISIM of any two nodes in the network is measured. Based on this new similarity and hierarchical clustering idea, the functional modules of the biological network are excavated, and the hierarchy of protein complexes and the multi-scale of functional blocks in biological networks are further revealed by the similarity of new nodes. We build a complete gene co expression network by fusion of multi condition gene coexpression data and analyze functional modules. Next, we break through the concept of modular structure as the only organizational form of functional units in a biological complex network, and a new functional module is found on the biological network. A two fork tree search method which can simultaneously excavate high aggregation and sparse function modules is proposed. Finally, some characteristics of the Bi-sparse module are deeply studied through statistical analysis of large-scale networks. Therefore, the research content and innovation points of this paper include the following aspects: (1) to be effective. In order to overcome the shortcomings of existing functional module mining methods, we use the restricted random walk model and propose a new transfer probability matrix, and then define a new node similarity ISIM to measure the distance between any two nodes in the network. The new node similarity has three good properties, one is that it can successfully fuse the network. The two is that the new node similarity can not only effectively measure the distance between the two nodes, but also capture the topology of the two nodes in the network; three is that it is in a convergent space to define the node similarity. Therefore, it is good in a series of incomplete and noisy biological networks. Stability and robustness. Using the new node similarity and hierarchical clustering idea, we can effectively analyze the functional modules in the biological complex network. First, we use the new node similarity to generate the similarity matrix of the network. Then, we use the hierarchical clustering idea to establish the tree structure of the network module. Finally, select the reasonable target function self. In this paper, a new method (isimb method) is proposed to reveal the multi-scale characteristics of the hierarchical structure and functional modules of the protein complex in the biological network. Compared with the existing module mining methods, the new node similarity is based on the new node similarity. The method is a non parametric method, it can automatically determine the number of modules in the network. The modules used to use it not only get better matching with the real functional module structure, but also can effectively overcome the defects of the incompleteness of the biological network. Compared with the single scale method, this paper innovatively introduces the concept of multi scale module to the module. To the mining of protein complexes and functional modules in biological networks, this new concept can not only successfully predict protein complex and its hierarchical characteristics, but also reveal the dynamic process of functional modules from a specific to general perspective. (2) this paper proposes that the incompleteness of the biological network and the non transmissibility of gene Co expression are proposed in this paper. A new method is used to detect the functional modules in the gene co expression network. This method first constructs a complete gene coexpression network with the gene coexpression data under different conditions, and then uses the maximum group algorithm to mine the functional modules in the network. This new method has a strong biological function compared with other methods. Similarity. Through the analysis of transcriptional and regulatory relationships, it is predicted that the high probability of genes in functional modules is regulated by the same transcription factor, thus providing rich results for the construction of gene regulation network. (3) the mining of functional modules in traditional biological networks is the only form of functional module organization based on the modular structure of high cohesion. However, this conclusion is doubtful in biological networks, especially in the protein interaction network. Therefore, we find a Bi-Sparse module different from the high aggregation module, and then combine the two forked tree theory and the matrix theory to propose a new method (BTS method) to mine the two types of functional modules.Bts method in the protein. The high aggregation modules and Bi-Sparse modules, which are excavated in the mass interaction network, make up functional units. Compared with other methods, the BTS method has good performance: first, it does not need to set the number of modules in advance; two the high aggregation module and the Bi-sparse module of the mining have significant biological function similarity. (4) we have high aggregation. Modules and Bi-sparse modules coexist in the same network as the organizational form of functional units. This concept is generalized. We collate four types of 25 networks and analyze the modules in the 25 networks by BTS method. The result is that (a) Bi-sparse module is universal; (b) in the social network, people in the Bi-sparse module are filled. In the role of broker, it is responsible for coordinating the contradictions among different groups and promoting the communication of information, technology and knowledge; in the computer software network, the nodes in the Bi-sparse module have similar software package properties; in the complex biological network, the protein or gene in the Bi-sparse module has significant functional similarity; (c) Bi-sparse modules in complex networks have some characteristics: first, compared to the Bi-sparse module and the high aggregation module, the Bi-sparse module contains less nodes; two, the Bi-sparse module has a certain preference in different types of networks; finally, there are two distinct topological structures in the Bi-sparse module in the complex network.

【学位授予单位】：上海交通大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：O157.5

【相似文献】