EpitopeVec: linear epitope prediction using deep protein sequence embeddings.

Area of Science:

• Immunoinformatics
• Computational Biology
• Vaccine Development

Background:

• B-cell epitopes (BCEs) are crucial for vaccines and diagnostics.
• Experimental BCE identification is costly and time-consuming.
• Existing computational methods lack generalizability, with cross-testing accuracies around 51-53%.

Purpose of the Study:

• To develop a novel computational method for rapid and accurate linear B-cell epitope prediction.
• To improve the generalizability of B-cell epitope prediction across different datasets and antigen types.

Main Methods:

• Developed EpitopeVec, a method utilizing residue properties, modified antigenicity scales, and protein language model-based representations (protein vectors).
• Benchmarked EpitopeVec against state-of-the-art methods on various datasets, including cross-testing scenarios.
• Trained a specialized linear viral B-cell epitope predictor using a large viral dataset.

Main Results:

• EpitopeVec demonstrated superior performance in accuracy and area under the curve compared to existing methods.
• Performance varied based on antigen species (viral, bacterial, eukaryotic).
• The dedicated viral BCE predictor showed improved cross-testing performance.

Conclusions:

• EpitopeVec offers a significant advancement in predicting linear B-cell epitopes.
• The method's improved generalizability and accuracy support its application in vaccine design and diagnostics.
• Specialized predictors, like the viral BCE predictor, can further enhance performance for specific antigen types.