Reliable B cell epitope predictions: impacts of method development and improved benchmarking

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• Antibody-antigen interactions are key to host defense against pathogens.
• B-cell epitopes are critical binding sites for antibodies, essential for vaccine design, diagnostics, and immunotherapeutics.
• Experimental epitope mapping is resource-intensive, driving the need for accurate in silico prediction methods.

Purpose of the Study:

• To address the underestimation of in silico B-cell epitope prediction performance.
• To present an updated DiscoTope method (DiscoTope-2.0) with improved epitope prediction capabilities.
• To demonstrate the impact of proper benchmark definitions on prediction accuracy.

Main Methods:

• Developed DiscoTope-2.0 incorporating a novel spatial neighborhood definition and half-sphere exposure surface measure.
• Evaluated DiscoTope-2.0 performance using cross-validation and independent datasets.
• Assessed the influence of refined benchmark definitions on prediction accuracy.

Main Results:

• DiscoTope-2.0 demonstrated improved performance compared to other state-of-the-art methods.
• Proper benchmark redefinition significantly enhanced average AUC performance from 0.791 to 0.824 on the training set.
• Average AUC performance on an independent evaluation dataset improved from 0.712 to 0.727.

Conclusions:

• Proper benchmark definitions are essential for accurately evaluating B-cell epitope prediction methods.
• DiscoTope-2.0 achieves highly significant predictive performance, offering a powerful tool for rational epitope discovery.
• The updated DiscoTope-2.0 method is available for broader research application.