An Improved Method for Predicting Linear B-cell Epitope Using Deep Maxout Networks

To establish a relation between an protein amino acid sequence and its tendencies to generate antibody response, and to investigate an improved in silico method for linear B-cell epitope (LBE) prediction. We present a sequence-based LBE predictor developed using deep maxout network (DMN) with dropout training techniques. A graphics processing unit (GPU) was used to reduce the training time of the model. A 10-fold cross-validation test on a large, non-redundant and experimentally verified dataset (Lbtope_Fixed_ non_redundant) was performed to evaluate the performance. DMN-LBE achieved an accuracy of 68.33% and an area under the receiver operating characteristic curve (AUC) of 0.743, outperforming other prediction methods in the field. A web server, DMN-LBE, of the improved prediction model has been provided for public free use. We anticipate that DMN-LBE will be beneficial to vaccine development, antibody production, disease diagnosis, and therapy.