WUREN: Whole-modal union representation for epitope prediction

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• B-cell epitope identification is crucial for developing vaccines, therapies, and diagnostics.
• Current molecular docking and AI methods for B-cell epitope prediction face limitations in accuracy, computational efficiency, and feature representation, especially for antigen-antibody complexes.
• There is a need for novel algorithms to improve protein information extraction for large-scale epitope prediction.

Purpose of the Study:

• To introduce WUREN (Whole-modal Union Representation for Epitope predictioN), a multimodal model designed for efficient B-cell epitope prediction.
• To address the limitations of existing methods in accurately representing protein features for epitope identification.
• To demonstrate the model's effectiveness in extracting comprehensive protein information.

Main Methods:

• Developed WUREN, a multimodal model integrating sequence, graph, and structural protein features.
• Evaluated WUREN on independent test proteins from the DiscoTope3 benchmark using both solved and AlphaFold-generated structures.
• Utilized AUC-PR scores to assess prediction performance.

Main Results:

• WUREN achieved AUC-PR scores of 0.213 on solved structures and 0.193 on AlphaFold-generated structures.
• The model demonstrated efficient feature extraction capabilities for protein complexes.
• The findings highlight WUREN's potential for accurate B-cell epitope prediction.

Conclusions:

• WUREN is an effective feature extraction model for protein complexes, advancing B-cell epitope prediction.
• The model shows generalizable potential for developing protein-based drugs and can be extended to other biomolecules like nucleic acids, carbohydrates, and lipids.