Font Size: a A A

Research And Application Of Protein-protein Interaction Prediction Based On Network Analysis

Posted on:2017-02-14Degree:MasterType:Thesis
Country:ChinaCandidate:Y Y ZhangFull Text:PDF
GTID:2180330488493973Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Protein-protein interactions play a very import role in many cell activities. It is of great significance through protein-protein interaction (PPI) prediction and functional studies to understand the process of life activities, disease prevention, research and development of new drugs. However, PPI prediction through biological experimental methods which will waste a lot of time and efforts, cannot meet the needs of rapid development of the life sciences. So PPI prediction through computational methods has become an important research topic of bioinformatics. With the development of high-throughput technology, people has obtained a lot of PPI data which makes it possible to predict protein interactions by the use of computational methods. In this thesis, PPI data viewed as complex network and furthermore make research on PPI prediction based on the principle of network theory.In this thesis, the main contents of PPI prediction research include about the following three aspects:(1) In this thesis, we proposed link prediction algorithm based on uncertain information dissemination map. We define the link information on each vertex pair according to the probability of links between the two nodes. The algorithm spreads the link information on the graph with a certain probability. On the graph, we set an edge weight between each pair of vertices to measure the ability to disseminate information between vertices. If the weights larger than others, the link information will be transmitted with larger probability. When a vertex pair received the link information from adjacent vertex, their original link information is retained in a certain proportion, and also be influenced by link information with a certain proportion transmitted from the adjacent vertices. This ratio depends on the vertices of the edge weights, and their adjacent vertex on the edge weights. When the dissemination process iterated until convergence,the link information between each vertex pair is the probability of the existence of links between them. We use a standard data set for testing, the experimental results show that the proposed algorithm has higher prediction accuracy and better statistical properties.(2) This thesis also presents community-based modularity PPI prediction algorithm. We take the PPI interactions as a complex network. In the network, the nodes represent proteins and edges represent interactions between proteins. Since each protein are likely to interact with other multiple proteins, so some regions in the network are dense, and some relatively sparse area. Newman is defined by the module to calculate the module of the community. The entire network is divided into one by one community. Through partition tree construction, the PPI network forms a hierarchical structure tree. In the tree, each leaf node represents a protein. The lower level of the lowest common ancestor between the protein pair, the more similar the two proteins. Moreover, the greater the likelihood of the same community the larger the possibility of protein pair being interacted. We calculate the similarity between protein pair to predict whether there are interactions. The algorithm can also simultaneously predict protein function.(3) In this thesis, we develop the online PIP system based on the two proposed prediction algorithms metioned above. It is able to predict the results in graphical form to show up.
Keywords/Search Tags:Protein-protein interaction, Information dissemination, Community modularity, Similarity, Partition tree
PDF Full Text Request
Related items