Predicting The Antigenicity Of Influenza A/H3N2 Virus Based On Deep Neural Networks

The A/H3N2 virus is a kind of respiratory virus,whose surface protein hemagglutinin(HA)is responsible for binding to cellular receptors,and causing viral invasion.And it is also the most important antigen for inducing host to produce protective antibodies,which is the main antigen of influenza vaccine.In order to escape host immunity,the amino acid sequence of HA changes very rapidly.This persistent and cumulative change will produce new antigenic strains,leading to seasonal flu epidemics and even global influenza outbreaks.Therefore,influenza viruses are a serious threat to human public health worldwide.At present,influenza vaccine is the most effective means of preventing influenza and preventing flu outbreaks.And the degree to which immunity induced by a vaccine strain is effective against another epidemic strain is mostly dependent on the antigenicity between the strains.Therefore,analysis of antigenicity is critical to the surveillance of influenza outbreaks and vaccine selection.Deep neural networks have been successfully applied to many fields,including bioinformatics.However,their effectiveness in predicting influenza antigenicity has not been tested.Therefore,aiming at the influenza A/H3N2 virus and its antigenic characteristics,this paper proposes a coding method for the antigenicity of the influenza A/H3N2 virus,designs prediction models for A/H3N2 antigenicity,conducts experiments to analyze the proposed model on two datasets.The experimental results show that the proposed models improve the performance of A/H3N2 antigenicity prediction.The specific work of this paper includes the following four aspects.Firstly,this paper proposes a coding method for the antigenicity of the influenza A/H3N2 virus.This code represents the amino acid and amino acid substitution information at each position on each virus pair as a vector.This coding method is not only free of manual setting but also not limited to certain sites.Therefore,the codes of A/H3N2 virus pairs lay a foundation for the prediction of antigenicity by deep learning.Secondly,based on bidirectional long short-term memory(BiLSTM),this paper design a prediction model for A/H3N2 influenza virus antigenicity.This model extracts the key features between the amino acids and amino acid substitutions in sequence pairs through a bidirectional LSTM layer,predict the antigenicity by a fully connected layer.Thirdly,based on convolutional neural network(CNN),this paper designs a prediction model for antigenicity of influenza A/H3N2 virus.This model extracts the interactions between the amino acids and amino acid substitutions by two convolutional layers.Fourthly,this paper applies convolutional neural networks combined with the attention mechanism to predict antigenicity of A/H3N2 influenza virus.The experimental results show that the prediction model can be further improved by increasing the depth and complexity of the model.