B-cell Epitope Prediction Method based on Deep Ensemble Architecture and Sequences

Predicting B-cell epitope is aiming the design of molecules that can mimic the structure and function of a genuine epitope and replace it in antibody diagnostics and therapeutics, as well as the design of the potentially safer vaccine. It is an essential step for immune therapy. Conformational B-cell epitope prediction based on antigen sequence is still a challenge for researchers. In this work, we construct a deep ensemble architecture for B-cell epitope prediction based on antigen sequences. We adopted one hot vector coding and physico-chemical properties schemes for encoding protein sequence fragments, and constructed seven independent convolutional neural network, then the weighted average method was used to integrate the seven networks. The proposed method is evaluated on the testing datasets of BepiPred 2.0. The experimental results show that the method achieves an AUC of 0.771, a sensitivity of 0.711, and a MCC of 0.222. In addition, we also evaluate the performance of our method on 13 independent testing cases and the results are superior to the existing methods. The comparisons show that our method is capable of predicting conformational B-cell epitopes based on a deep ensemble framework and antigen sequences in an acceptable level of AUC. The codes along with instructions to reproduce this work are available from https://github.com/yangycoding/DeepBcellPre.