Research On Function Modularity Analysis And Module Recognizing Methods For Biological Networks

In recent years, massive amounts of data produced by high-throughput technologies make systems biology become a hot research area gradually. As the abstract representation of biological systems, biological networks play a crucial role in revealing biological processes (or functional) implementation mechanisms. The analysis of their organization is the most direct way to understand the relationship between network structure and their functions. Therefore, functional modularity analysis of biological networks is one of the core problems in systems biology research.A large number of studies have indicated that most of biological network show functional modularity, what is actually a conclusion derived from most biological networks. However, the futher study found that the relationships between the modules identified by these methods and biological functions are complicated and there is no clear corresponding connection between them. We need an evaluation system to measure functional modularity of network systematically and to display a more accurate view of the whole network. In addtion, identifying condition-specific modules is more useful to discribe the dynamic feature of networks, and condition-specific modules is suitable for analyzing networks with poor functional modularity.In view of these problems, our works presented the methods to define functional modules firstly, and then defined a serial of metrics for evaluating the functional modularity of biological networks. Moreover, we developed a criterion to judge whether a network has functional modularity or not. Applying these metrics and criteria on transcription factor regulatory networks, protein-protein interaction networks and metabolic networks, results showed several significant conclusions. In addition, according the characters of functional modularity in transcriptional regulatory network and its dynamic property, we proposed a method to identify condition-specific regulatory network modules. Finally, we summarized our works and innovations as"one index system","two biological significant discovery"and"three algorithms or improvement". More specific introductions are described below.1) Refering to the modularity notion in software engineering field, we defined a new metric to evaluate the functional modularity of biological networks, and developed a criteria to judge whether a network have functional modularity or not. The index system consists of three indexes: cohesive degree for each module, coupling degree between each pair of modules and MCC (Modularity based on Cohesive and Coupling). Comparing with other modularity index, they were more suitable for the measurement of biological networks, and can be used in the analysis of both directed graphs and undirected ones.2) We used the index systerm and the criteria to analyze transcription factor regulatory networks and protein-protein interaction networks in yeast, the results indicated that the transcription factor regulatory network was not of functional modularity, whereas protein-protein interaction networks shows strong functional modularity. That means not all types of biological networks have good functional modularity. After a comparative analysis of these two networks, we believed that the functional specificity of biological molecules in network affects the existence of the network functional modularity. Transcription factors are widely engaged in the cell function implementss, their functional specificities are significantly lower than that of protein-protein interaction network. This resulted directly in the leak of functional modularity of transcription factor regulatory networks.3) Considering the functional modularity of the metabolic network enzyme graph in 23 eukaryotes, 136 bacterial species and 18 species of archaea, we found that all metabolic networks in these species showed significant functional modularity. More importantly, we analyzed the relationship between the functional modularity of these metabolic networks and their evolution distance in phylogeny tree. The functional modularity of species in eukaryotes and bacterias with longer evolutionary distances showed greater functional modularity, while the archaea species show contrast results. That is because most of the archaea species live in an extreme environment, which leads to a smaller evolutionary pressure from environment change.4) The definition methods of functional modules are very important for the analysis of functional modularity in biological networks. As algorithms for the definition of functional modules still have some problems, we proposed serveral basic principles of defining directly the functional modules. In the applications of these basic principles, we use MIPS functional annotations to define the functional modules of yeast transcriptional regulatory network. A large number of functional modules are discrete which means the functional modularity is poor. In addition, we program the functional module definition method based on GO functional annotations into Cytoscape plug-in using this software, we displayed the distribution of the functional modules in yeast transcription factor regulatory network by visualization tools.5) In calculation of the functional modularity, we need to find the shortest path between all pairs of nodes whose elements are in different functional modules. This is a task with high computational complexity. Firstly, we calculated the coupling degree of each pair of functional modules using traditional minimum path algorithm and then obtained the modularity (we call this process as"the traditional method"). As there are many double-counting in the calculation by the traditional method, we proposed a fast calculation method, called DNM algorithm. The algorithm could handle all these special pairs of nodes in ONE loop, and the computational complexity is lower.6) As transcription factors carried a large amount of functions in the transcriptional regulatory function, yeast static transcriptional regulation network shows no significant functional modularity. Therefore, condition-specific dynamic regulatory modules to a certain extent represent the functional modules in transcriptional regulatory network. Using the time series gene expression microarray data under the cell cycle, we modeled gene's transcription regulation process by differential equation system. And we solved the equation from a new perspective (treating the differential equation model as an optimization problem and solved it by using genetic algorithms, rather than solved it by using the traditional methods of differential equations). The results showed that the solving process for the majority of target genes is convergent. Validations proved that our algorithm shows higher performance than the existing methods. In addition, we obtained two interesting results: the correlation among the co-regulated transcription factor in condition-specific modules is closer than that in static network (namely the entire regulatory network); there are many potential regulatory complexes in cell cycle.In a word, we focus on the research on following aspects: defining the function modules, bringing up forward metrics about functional modularity, analyzing the functional modularity of several types of biological networks, recognizing the condition-specific transcriptional regulatory sub-networks. These analysis will help to view the functional modularity of biological networks.