Research And Implementation Of Virus-host Protein-protein Interaction Prediction System Based On Fusion Sequences And Network Embedding

At present,infectious diseases caused by the virus infection have brought unprecedented challenges to the global medical system,seriously threatened the health of people all over the world,interfered with people's daily life,and caused heavy economic losses.Virus transport their own genetic material into the host cell nucleus through interspecific interactions of proteins and RNA,controlling cellular metabolism and disrupting host cell functions.Therefore,identifying protein-protein interactions(PPI)between virus and host is a key point for deciphering the complex virus-host relationship.Rapid identification of virus-host protein-protein interactions is of great significance for understanding the host immune response mechanism,finding new drug targets,and developing vaccines.The experimental identification of traditional PPI is often time and economical,and it is difficult to obtain complete protein interactions.However,the prediction of PPI by computing technology has the advantages of low cost and high success rate.Therefore,in recent years,the technology of predicting PPI by computer aids has developed rapidly.It has been found that host proteins targeted by viruses usually have special topological properties in the host protein-protein interaction network.In response to this point,this experiment constructed a virus protein sequence similarity network and a host protein-protein interaction network.At the same time,this experiment fused the protein sequence features and network features and used a long short-term memory network(LSTM)to build the model.The experiment first used the Smith-Waterman algorithm to calculate the local sequence similarity of viral proteins to complete the construction of the viral protein sequence similarity network and constructed the host's protein interaction network based on the host's PPI data.Then,the protein sequences are encoded with features,and the Doc2Vec algorithm is used to obtain the protein sequence feature encoding of the virus and the host,and the node2vec algorithm is used to obtain the virus and host protein network feature encoding.Sequence and network feature encoding of virus and host are fused as input to subsequent models.Finally,the LSTM model is trained to make predictions on the potential proteins and proteins of the virus and host.Based on the above experiments,a virus-host protein-protein interaction prediction system is developed in this thesis.The system is divided into a user end and a management end.On the client-side,after logging in,users can predict the interaction between virus-host proteins and proteins,visualize the prediction results on the network,view and download the prediction results,manage personal information,and download public resources.On the management side,after logging in,the administrator can perform user information management,result management,and common dataset resource management.The front-end of the system adopts the Bootstrap framework,and the front-end and back-end interaction functions are completed using Python and PHP languages.