| In the post-genomic era, biological network as a complex network, has receivedextensive attention. In order to comprehensively understand how interactions betweenproteins to complete the life activities, we have to analyze the characteristics of theprotein-protein interaction(PPI) network and through these characteristics mining proteincomplexes (also called modules) and predicting the functions of unknown proteins.In2006, Gavin et al. found that the protein complex consists of a core and attachmentproteins. Recently, many computational methods for identifying the core-attachmentcomplexes have been proposed. In this paper, we mainly study this kind of algorithms.Firstly, we improved the CoAch (COre-AttaCHment) algorithm. According to thecharacteristics of PPI networks, a new rule for adding attachment proteins is presentedand the algorithm is optimized. Then the idea of random walk is applied to find the cores ofcore-attachment protein complexes. Comparison with other different algorithms for miningcomplexes in PPI network, we found that the improved algorithm is more accurate inpredicting protein complexes.Finally, we make a study on a spectral clustering method which is used for partitioningcomplex networks. The traditional spectral clustering method must be pre-determined thenumber of clusters and for large-scale data its time complexity is high. Therefore, weimprove the traditional clustering algorithm by adding a preprocessing step. Also, in thealgorithm we use the modularity Q as a measure of the quality of the network partition. Theexperimental results show that our new method improves the accuracy of the networkpartition and can process the large-scale complex networks with less time cost.In conclusion, though the three algorithms studied in this paper still need to beimproved in many aspects, they all improve the performance of the original algorithms andthey have their own application advantages compared with existing similar algorithms. |
PDF Full Text Request