SEMA: Antigen B-cell conformational epitope prediction using deep transfer learning.

Area of Science:

• Computational biology
• Immunoinformatics
• Structural biology

Background:

• Accurate prediction of conformational B-cell epitopes is essential for developing effective vaccines and immunotherapies.
• Current methods for epitope prediction often have limited accuracy across diverse antigens.
• Identifying antibody binding sites on antigen tertiary structures remains a significant challenge.

Purpose of the Study:

• To develop a novel deep learning model for predicting conformational B-cell epitopes.
• To leverage transfer learning with pretrained protein language models for improved epitope prediction accuracy.
• To provide a tool that utilizes both antigen primary sequence and tertiary structure for prediction.

Main Methods:

• Applied transfer learning using pretrained deep learning models (ESM-1v, ESM-IF1).
• Fine-tuned models to predict antibody-antigen interaction features and differentiate epitope from non-epitope residues.
• Developed the SEMA model integrating sequence and structural information.

Main Results:

• The SEMA model achieved a ROC AUC of 0.76 on an independent test set, outperforming existing peer-reviewed tools.
• Demonstrated SEMA's capability to quantitatively rank immunodominant regions within the SARS-CoV-2 RBD.
• Validated the model's performance on predicting key antigenic sites.

Conclusions:

• SEMA offers a significant advancement in conformational B-cell epitope prediction accuracy.
• The model's ability to integrate sequence and structural data enhances its applicability in vaccine design.
• SEMA provides a valuable resource for immunoinformatics and drug development, with code and a web interface available.