Abstract: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 antibody-specified 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 antigen-antibody 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.

Predicting the distance between antibody’s interface residue and antigen to recognize antigen types by support vector machine

Prediction of Functional Class of Proteins and Peptides Irrespective of Sequence Homology by Support Vector Machines.

Comparison study of two kernel-based learning algorithms for predicting the distance range between antibody interface residues and antigen surface

Prediction of nucleic acid-binding proteins using support vector machines

Machine learning prediction of Antibody-Antigen binding: dataset, method and testing

PREDICTION OF MHC CLASS-II BINDING PEPTIDES BY SUPPORT VECTOR MACHINES

AbAgIntPre: A Deep Learning Method for Predicting Antibody-Antigen Interactions Based on Sequence Information

Prediction of MHC-binding Peptides of Flexible Lengths from Sequence-Derived Structural and Physicochemical Properties

Exploiting the Role of Features for Antigens-Antibodies Interaction Site Prediction

Computer Prediction of Allergen Proteins from Sequence-Derived Protein Structural and Physicochemical Properties

MVSF-AB: Accurate antibody-antigen binding affinity prediction via multi-view sequence feature learning

QSAR method for prediction of protein-peptide binding affinity: application to MHC class I molecule HLA-A*0201.

Prediction of HLA-DRB1*0401 binding peptides using support vector machine

Prediction of Rna-Binding Proteins from Primary Sequence by A Support Vector Machine Approach

HAIRpred: Prediction of human antibody interacting residues in an antigen from its primary structure

Prediction of T-cell epitopes based on least squares support vector machines and amino acid properties.

Predicting Rrna-, Rna-, and Dna-Binding Proteins from Primary Structure with Support Vector Machines

Using a Machine-Learning Approach to Predict Discontinuous Antibody-Specific B-Cell Epitopes

Computationally Probing Drug-Protein Interactions Via Support Vector Machine

Homology-Free Prediction of Functional Class of Proteins and Peptides by Support Vector Machines

Function Prediction for Dna-/Rna-Binding Proteins, Gpcrs, and Drug Adme-Associated Proteins by Svm