Research On Key Protein Prediction Methods Based On Network Topology And Biological Information Fusion

Essential proteins are vital material and functional foundations for life activities in living organisms.The normal viability and fertility of living organisms are inseparable from the participation of essential proteins.The identification of essential proteins,not only contributes to the understanding of biological mechanisms,but also provides significant guidance in the research of disease diagnosis and drug targets.In recent years,with the rapid development of information technology,predicting essential proteins from protein-protein interaction(PPI)networks by using computational biology techniques has become an important proteomics research.In the previous researches,experimental technologies are the main approaches for the interactions between proteins.With the continuous development of computer science and technology,using computational technologies to predict and analyze PPI networks increasingly becomes a new mainstream research direction of bioinformatics.However,due to the complexity of the intracellular environment and the limitations of related technical means,usually,the obtained PPI data are highly false-positive and false-negative.Moreover,the interactions between proteins are dynamically changing over time,environments and different stages of cell cycle.Therefore,PPI data can not reflect the real protein-protein interactions in cells,which seriously hampers the accuracy of discovering essential proteins from PPI networks.In this paper,in order to improve prediction accuracy,adopting random walk model as well as intelligent optimization algorithms for protein essentiality prediction on the basis of an efficient fusion of the topological and biological properties of proteins.The main research work of this thesis includes:Firstly,predicting essential proteins based on participation degree in protein complex and subgraph density.First,construct relatively reliable PPI networks.Then,integrate the topological and biological properties of proteins by means of three information fusions.Finally,predict essential proteins according to the final scores.Simulation results show that the new proposed centrality method based on information fusion can accurately predict more essential proteins than traditional centrality methods.Secondly,predicting essential proteins based on random walk with restart model.First,construct a weighted PPI network by using Edge clustering Coefficient(ECC),gene expression and GO annotation information.Then assign initial scores for proteins by using subcellular localization and protein complex information.Subsequently,perform the random walk with restart algorithm on the weighted PPI network,and the final ranking scores can be achieved when the algorithm reaches its termination.Finally,predict essential proteins according to the ranking scores.Simulation results show that the new proposed method can obtain a superior prediction performance compared to other state-of-the-art essential proteins identification methods.Thirdly,predicting essential proteins by applying the optimizing mechanism of artificial fish swarm algorithm.First,initialize artificial fishes by utilizing known essential proteins,and take the known essential proteins as priori knowledge for essential protein prediction.Then the foraging behavior,random behavior,swarming behavior and following behavior are performed to predict essential proteins.Simulation results reveal that the novel method achieves superior results for identifying essential proteins in comparison with several traditional competing identification methods.Finally,predicting essential proteins by applying the optimizing mechanism of immune algorithm.A candidate set of essential proteins are viewed as an antibody.Updating the antibody population by means of genetic operators and immune operators.The predicted essential proteins can be obtained when the algorithm reaches its termination.Simulation results show that the performance of the new algorithm outperforms other competing methods on the prediction of essential proteins.