Background: Predicting B-cell epitopes is important for understanding disease pathogenesis,
identifying potential autoantigens, and designing vaccines and immune-based cancer therapies. The
experimental approaches used for detecting B-cell epitopes are often laborious and resource-intensive.
Thus, several computational methods have been developed for predicting the epitopes of a given
antigen. However, most of these methods are coarse binary classifications of antigen regions within
epitopes or non-epitopes and do not specify antibodies. Therefore, we aim to solve this antibodyspecified
epitope prediction problem using a developed structure-based computational machine-learning
method, Epitopia, to reflect this biological reality accurately.
Result: We selected 60 non-redundant antibody-antigen protein complexes to train and applied the
leave-one-out cross-validation method to test the accuracy of our proposed methods; we compared the
results with the Epitopia, Discotope 2.0,PEASE and the state-of-the-art tool SEPPA 2.0. We considered
the role of both complementarity determining region residues and antigen surface residues in antigenantibody
interactions and assigned a score for each antigen surface residue. If we considered a
prediction to be successful if the average score of “epitope residues” exceeded the average score over all
surface residues. Then the success rates of our proposed methods were all higher than Epitopia and the
best one was 83.3%, which is 7% higher than the rate obtained with Epitopia (76%). The results show
that antibody-specific methods are competitive with, and sometimes even better than, Epitopia, which
only considers the antigen residues.
Conclusion: Our antibody-specific methods provide sufficient accuracy in locating epitopes because
whether an antigen surface residue is an epitope depends on whether the antibody recognizes it. This
approach is a new method for identifying B-cell epitopes. Based on our results, we believe that the
prediction of antibody-specific B cell epitopes is more meaningful and efficient than those which are
only considering antigen residues.