Functional Module Mining In Biological Complex Networks

Network science, which is an emerging interdiscipline, is widely studied, and its fundamental theories are closed related to various fields, such as mathematical science, biological science, engineering science and social science. Many scientists come from different disciplines have been paid more attentions to complex networks. Recognizing and understanding the qualitative and quantitative properties of complex networks is a significant and challenging project. One of the most relevant and ubiquitous features of complex networks is module or community structure. Detection and analysis of module structure has theoretical and practical significance for understanding evolution, structure and dynamics of complex networks.As organization of functional modules in biological networks, module structure has significant meaning and application in life sicence. Therefore, many effective algorithms have been proposed to detect modules from complex networks, such as graph-theoretic algorithms, the algorithms based on random walk model and spectrum clustering. But most of these methods suffer from drawbacks which come from the algorithm itself and the imcomplete and noise of biological networks. Therefore, we should focus special study on these issues and put forward to unorthodox methods.In this dissertation, the study focuses on how to detect functional modules and analyze their organization in biological complex networks. First, in order to solve the defects of current algorithms, we propose a new node similarity(named ISIM) in a network, and then detect modules by combining the new similarity and hierarchical clustering. At the same time, both hierarchical structure of protein complexes and multi-scale functional modules in biological networks are revealed by the new node similarity. Second, in order to conquer the limitation of imcomplete, we combine multi-conditional gene co-expressed data to construct complete gene co-expressed networks, and then detect functional modules. Third, we break away from the traditional concept that cohesive module structure is unique organization of functional units in biological complex networks, and we proposed a new method, BinTree Seeking(BTS) method, to mine function modules in biological networks, which are constructed by cohesive and Bi-sparse modules. At last, some features of Bi-sparse module are studied by a large-scale analysis on 25 complex networks. The main contents and creations of this dissertation are summarized as follows:(1) In order to overcome the drawbacks of traditional module detection methods, we propose a new node similarity, named ISIM, using constrained random walk and motified transition probability matrix to detect modules. The new node similarity has three excellent merits. The first one is that it can capture both global and local network topology information successfully. The second one is that the new node similarity not only defines the distance between two nodes but also describes their topology structures profoundly. At last, the proposed ISIM does not work on the observed data directly, but measuring the similarity between any pair of nodes in a mapped convergent space. This feature will make the ISIM robust to noisy and incomplete biological networks.The new node similarity is then applied to detect functional modules in complex networks. First, we use the new node similarity to generate a similarity matrix of input network. And then a dendrogram can be built by hierarchical clustering. At last, the tree is automatically cut by the partition density and the module structure of network is generated. Furthermore, by adjusting the balanced factor in the new node similarity, we introduce an algorithm ISIMB to reveal both hierarchical organization of protein complexes and multi-scale functional modules in biological networks.The ISIM method is parameter-free method, and can select the number of modules automatically. The predicted functional modules can better match with reference module structure than other methods. ISIM method can also overcome the influence of incomplete on accuracy. In contrary to single-scale methods, the concept of multi-scale module is brought into biological networks, and ISIMB method opens an unorthodox way to study functional modules: from specific to general.(2) Contraposing the imcomplete biological networks and intransitivity of gene co-expression, we propose a new method to mine functional modules in gene co-expressed networks. The new method first constructs a integrate gene co-expressed network using multi-conditional gene co-expressed data, and then detects functional modules with maximum clique algorithm. Comparing with other methods and pathway data, the predicted results from the new method shows better performace on biological functions. By analyzing the transcription factor(TF)-gene relationship, we find that the genes in the predicted functional modules are regulated by the same transcription factor with a high probability. The phenomenon can provide abundant results to construct gene regulatory networks.(3) Most functional module detection algorithms depend on the traditional hypothesis that a functional module in biological networks is a cohesively linked group of nodes, densely connected internally, and sparsely interacting with the rest of the network. However, recent studies have revealed that it is not always the case that members in the functional module link each other densely in biological networks. Therefore, a new functional module structure called Bi-sparse module is first defined in this work, and then BinTree Seeking(BTS) method is proposed to detect both Bi-sparse and traditional cohesive modules in protein-protein interaction(PPI) networks. In contrary to other methods, BTS method has advantages that it can automatically find optimized number of modules in large PPI networks, and that the mined groups including cohesive and Bi-sparse modules have significantly related biological functions.(4) We apply the general principle that Bi-sparse modules co-exist with cohesive modules in the same complex network to broader types of networks. Experiments on 25 complex networks, which can be indivied into four types, show that:(a) Bi-sparse modules commonly exist with cohesive modules in complex networks;(b) the people in Bi-sparse modules promote information, goods, opportunities, or knowledge flow across different groups in social networks; the nodes in Bi-sparse modules share similar attributes in computer software networks; the proteins and genes in PPI networks and gene co-expressed networks also have similar biological functions;(c) some characteristics of the Bi-sparse modules are observed: generally, Bi-sparse modules are smaller than cohesive modules in the same network; Bi-sparse modules appear ubiquitous in nature, but have a preference for some types of networks; At last, two possible organizing structures for Bi-sparse modules in complex networks are observed from the results on the large-scale networks.