Modular Approach Of Protein Networks

This thesis provides a new definition of similarity for clustering of protein-protein interaction networks, and obtaining valid modules. In order to better define the similarity, we first obtain the protein-protein interaction datasets of multiple species, and gather statistics of the distribution of protein-protein interaction numbers on the second-order and the third-order interaction number of paths. And, then compare it with random case. The result shows that the distribution on the second-order has significant shift relative to the random. However, the distribution on the third-order tends to random. This indicates that the second-order interactions in the protein-protein interaction network better reflects the similarity of proteins, and has more biological significance. Based on this conclusion, we define a more reasonable similarity, the coefficients of which can be adjusted according to different model organisms to make it more purposiveness. Finally, we modularized Saccharomyces cerevisiae protein interaction network according to fuzzy clustering algorithm based on this similarity definition, and evaluate the clustering results which were proved to be good modules.