Study On The Prediction Of B Cell Epitopes Based On LSTM Network With Multiple Source Information

In recent years,with the deepening of biological research,epitope prediction has become a hot topic of research.When the body is stimulated by external antigens,it can secrete antigens,stimulate B cells and cause humoral immune response.Epitopes play an important role in the immune process.B cell epitopes are fragments of an antigen that are recognized by and react with B cell antigen receptors.Experimental methods for B epitope prediction can get more accurate results,but it takes a long time and consumes manpower,material resources and financial rescources at high costs.Therefore,the use of computer B-cell epitope prediction method for its low-cost,highspeed features have been widely used.In order to improve the effect of epitope prediction,in this paper,we take the epitope data in the IEDB database as the research object,and propose an epitope prediction model based on LSTM network with multi-source information.We select three kinds of feature information of amino acids for epitope prediction in this model.By using the proposed prediction model to analyze the two datasets,it is possible to exploit the different characteristics between epitopes and non-epitopes and we can take it as the theoretical foundation and basis for the development of vaccines.The main work of this paper is as follows.(1)The method of features extraction with multi-source is presented.This method digitally represents three kinds of characteristic information derived from B-cell epitope fragments,analyzes the influence of these characteristics on the epitope prediction,and the ability to discriminate between positive and negative samples of antigenic epitopes.At last,it indicates the validity of feature information.(2)This thesis proposes a B cell epitope prediction model based on LSTM network with multi-source information.The model firstly integrates three kinds of acquired digital characteristic information,uses principal component analysis method to reduce dimension for fusion features and builds a prediction model based on LSTM network.On the two data sets,three comparison models including SVM,recurrent neural network and LSTM network are studied.The results show that the proposed prediction model incorporates a variety of epitope features,increases the amount of information in the feature and makes use of the method of dimensionality reduction to reduce the redundant information,which improves the accuracy of the prediction model.(3)The proposed LSTM prediction model is used to predict the prion protein sequence PRNP which is marked epitopes in fact.The comparison between the predicted results and the labeled results further demonstrates the prediction performance of the proposed model.