MsgaBpred: A B-cell epitope predictor integrating AlphaFold3-predicted structures with multi-scale GCNs and...

Area of Science:

• Computational biology
• Immunoinformatics
• Structural bioinformatics

Background:

• Accurate B-cell epitope prediction is crucial for developing vaccines, therapeutics, and diagnostics.
• Experimental methods for epitope identification are time-consuming and costly.
• Current computational methods often depend on experimentally resolved or low-accuracy predicted protein structures, limiting their effectiveness.

Purpose of the Study:

• To develop a computational model, MsgaBpred, for B-cell epitope identification utilizing AlphaFold3-predicted protein structures.
• To overcome limitations of existing methods by not requiring experimentally determined structures.
• To improve the accuracy and efficiency of B-cell epitope prediction.

Main Methods:

• MsgaBpred employs a multi-scale graph convolutional network and additive attention mechanism.
• The model processes protein sequences to capture complex 3D structural dependencies.
• It utilizes ESM-C, an advanced protein language model, for enhanced feature representation.

Main Results:

• MsgaBpred demonstrates competitive and robust performance across multiple benchmark datasets.
• The model achieves a statistically significant improvement in Area Under the Curve (AUC) compared to state-of-the-art methods.
• The multi-scale design effectively models both local and global structural contexts.

Conclusions:

• MsgaBpred offers an efficient and accurate computational strategy for B-cell epitope prediction using predicted protein structures.
• The model's performance surpasses existing state-of-the-art methods, particularly in AUC.
• Its modular and scalable architecture suggests potential for broader applications in biomolecular structure analysis.