Research On Protein Complexes Detection Based On Deep Autoencoder

With the continuous development of bioinformatics,research on protein complexes is of great significance for exploring the mysteries of cells and life.Traditional bioassay-based complexes detection methods have many shortcomings.In recent years,with the development of high-throughput technology,computation-based protein complexes detection methods have gradually become the mainstream.Most of the existing protein complexes detection methods utilize traditional clustering algorithms on protein-protein interaction(PPI)networks.However,due to the complexity of the network structure,such a direct manner cannot fully utilize the information contained in the network.As a new data preprocessing method,network embedding can effectively extract network information,thus improving the quality of analysis.Therefore,based on the idea of network embedding,this paper proposes a novel protein complexes detection method called deep attributed network embedding(DANE).Firstly,the method extracts network information by using the deep autoencoder,and adds the biological function information as an auxiliary on the basis of the network structure information,then obtains the vector representations of the nodes.The proteins are then clustered using the maximal cliques generating algorithm based on the core-attachment structure to obtain the final protein complexes prediction.The deep autoencoder framework used in this method can effectively extract the nonlinear information and filter the noise interference contained in the data.Based on the DANE method,this paper proposes a novel protein complexes detection method called neighbor similarity attributed network embedding(NANE),which focuses on the global structure information of the network.The method calculates the similarity of the nodes in the neighbor structure through the normative indicators,and reconstructs the loss function.Through the effective extraction of the global structure information,the performance of NANE is further improved.In this paper,comparative experiments were carried out on six PPI datasets to verify the effectiveness of the two methods DANE and NANE.The experimental results show that the two methods have better performance than the existing methods in various evaluation indicators and have high availability.