Structure-based Predictions of Conformational B Cell Epitopes by Protein Language Model and Deep Learning.

Area of Science:

• Immunology
• Computational Biology
• Structural Biology

Background:

• Mapping B-cell epitopes is crucial for antibody discovery but remains challenging due to experimental costs and limitations of current computational tools.
• Existing computational methods often struggle with antibody-antigen recognition, performing poorly on specific interfaces.
• Accurate epitope prediction is vital for advancing antibody engineering and vaccine development.

Purpose of the Study:

• To introduce a novel patch-centric computational framework for predicting B-cell epitopes directly on antigen structures.
• To evaluate the performance of a protein language model (PLM) approach against a convolutional neural network (CNN) baseline for epitope prediction.
• To develop an interpretable method for identifying and prioritizing B-cell epitopes to aid in antibody discovery and vaccine design.

Main Methods:

• Developed a patch-centric framework defining epitopes as triads of neighboring residues capturing local shape and chemistry.
• Evaluated two classifiers: a PLM (ESM-2 embeddings) and a CNN using hand-crafted features.
• Trained and validated models on 1,151 antibody-antigen complexes from the AbDb database using five-fold cross-validation.

Main Results:

• The PLM classifier significantly outperformed the CNN at the patch level (F1 ≈ 0.986, ROC-AUC ≈ 0.998).
• Residue-wise performance aggregated from patch scores showed superior results for the PLM (ROC-AUC 0.689±0.072 vs. 0.548±0.018).
• The PLM achieved state-of-the-art performance on benchmark datasets and generalized well to external complexes (ROC-AUC 0.663).

Conclusions:

• The patch-centric PLM framework offers a robust and accurate method for B-cell epitope prediction.
• The model provides interpretable epitope likelihood maps, facilitating antigen prioritization and antibody engineering.
• This approach represents a practical advancement for antibody discovery, vaccine design, and related biotechnological applications.