BepiPred-2.0: improving sequence-based B-cell epitope prediction using conformational epitopes

Abstract

Antibodies have become an indispensable tool for many biotechnological and clinical applications. They bind their molecular target (antigen) by recognizing a portion of its structure (epitope) in a highly specific manner. The ability to predict epitopes from antigen sequences alone is a complex task. Despite substantial effort, limited advancement has been achieved over the last decade in the accuracy of epitope prediction methods, especially for those that rely on the sequence of the antigen only. Here, we present BepiPred-2.0 (http://www.cbs.dtu.dk/services/BepiPred/), a web server for predicting B-cell epitopes from antigen sequences. BepiPred-2.0 is based on a random forest algorithm trained on epitopes annotated from antibody-antigen protein structures. This new method was found to outperform other available tools for sequence-based epitope prediction both on epitope data derived from solved 3D structures, and on a large collection of linear epitopes downloaded from the IEDB database. The method displays results in a user-friendly and informative way, both for computer-savvy and non-expert users. We believe that BepiPred-2.0 will be a valuable tool for the bioinformatics and immunology community.

Figure 1.

The Summary output page in Advanced Output mode, showing BepiPred-2.0 and NetSurfP predictions for each query sequence.

Figure 2.

A heat map of each feature's impact on the 5 models generated from the 5-fold cross validation. Ranging from yellow to red, where yellow is low impact and red high impact. Each row specifies one cross-fold RF model and the columns specify the feature. The ‘.Nx’ and ‘.Cx’ suffix specifies positions relative to the investigated residue towards N and C-terminals, respectively.

Figure 3.

The average predictive positive value (PPV) (A) and true positive rate (TPR) (B) across all antigen sequences in test set using different number of top scoring residues. Four different methods are evaluated: BepiPred 1.0 (red), NetSurfP (black), LBtope (gray) and BepiPred 2.0 (blue).

Figure 4.

Lysozyme (displayed as surface, coloured from blue to red according to BepiPred-2.0 predictions) with four unique epitope regions obtained from antibodies 1BVK (purple), 1C08 (red), 1MLC (yellow) and 4TSB (green).