Research Of Protein Complex Extraction Based On Protein-Protein Interaction Network

Protein complex is a macromolecular biological structure consisted of two or more associated proteins formed by interactions. And the proteins in the same protein complex perform biological function by cooperating with each other. Protein complex plays a critical role in many biological processes, so it is fundamental to the understanding of the principles of cellular organizations and biological functions. As a respond to wildly used of high throughtput approach involve genome-wide detection of protein interactions, the volume of protein-protein interactions is expanding at an incredible rate, which can help provide a brand new method to predict protein complexes. Therefore, it is a hot research area to utilize the protein-protein interactions to extract protein complexes.In this paper, we first introduce the related knowledge and present research summary of extracting protein complexes. Then, we present the evaluation metrics of predicting protein complexes. And we propose an effective method to extract protein complexes from PIN.First of all, an algorithm based on iterating revised protein interaction network is proposed. The method utilizes the topological characteristics to compute the weight of the edges in the network, and filtering the noisy interactions and adding high reliability interactions into the network. Then we extract all the maximal sub-graph from the revised network, filtering and merging the sub-graph with highly overlapped, the left sub-graphs are the predicting complexes.In order to exploit the biological structure and functional properties of protein complexes, the algorithm incorporates the functional annotations into the detection of protein complexes. The function annotations are employed to compute the function similarity of proteins in the network, and processing the protein interaction network based on the similarity. According to the core-attachment structure of protein complexes, the method extracts the core proteins of the complexes and adds the attachment proteins into the clusters to form protein complexes.Finally, this paper optimizes the algorithm according to the existing problems with the prediction of protein complexes from protein interaction network. Not only utilizing the protein functional annotations in the step of computing protein functional similarity, but also in the process of filtering non-qualified maximal sub-graphs. What's more, the machine learning method of SVM is introduced into this method. The algorithm selects proper feathers and classifier to filter the extracted complexes to improve the precision of the prediction of protein complexes.In conclusion, according to the topological characteristics of protein interaction network and the functional properties of protein complexes, we proposed a competitive algorithm to extract protein complexes from protein-protein interactions. We have applied our methods successfully on several protein-protein interaction databases and protein complexes databases, and achieve competitive results.