Identifying B-cell epitopes using AlphaFold2 predicted structures and pretrained language model.

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• B-cell epitope identification is crucial for designing vaccines and immunotherapies.
• Experimental methods for epitope prediction are costly and time-intensive.
• Existing computational methods often neglect structural information, limiting their accuracy.

Purpose of the Study:

• To develop a novel, accurate computational model for B-cell epitope prediction.
• To leverage protein structural information for improved epitope prediction.
• To provide a freely accessible web server and resources for researchers.

Main Methods:

• Proposed GraphBepi, a graph-based model utilizing AlphaFold2-predicted protein structures.
• Encoded protein residues using ESM-2 representations within a graph structure.
• Employed an edge-enhanced deep graph neural network (EGNN) to capture spatial information.
• Integrated EGNN with bidirectional long short-term memory networks (BiLSTM) for sequence dependencies.
• Combined EGNN and BiLSTM features in a multilayer perceptron for epitope prediction.

Main Results:

• GraphBepi demonstrated superior performance compared to state-of-the-art methods.
• Achieved improvements of over 5.5% in Area Under the Curve (AUC).
• Achieved improvements of over 44.0% in Area Under the Precision-Recall Curve (AUPR).

Conclusions:

• GraphBepi offers a significant advancement in B-cell epitope prediction accuracy.
• The model effectively integrates structural and sequential information for enhanced performance.
• The developed web server and open-source code facilitate broader research applications.