BLMPred: Predicting linear B-cell epitopes using pre-trained protein language models and machine learning

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• B-cell epitopes are key for immune responses, crucial for diagnostics, vaccines, and immunotherapy.
• Experimental epitope identification is costly and slow, necessitating efficient computational methods.

Purpose of the Study:

• To introduce BLMPred, a novel computational tool for predicting linear B-cell epitopes.
• To leverage protein language model embeddings for epitope prediction without 3D structures.

Main Methods:

• BLMPred employs a sequence-based approach using pre-trained protein language model embeddings.
• It functions as a binary classifier to predict epitope presence in peptide sequences.
• The method derives local and global protein structural features from the primary amino acid sequence.

Main Results:

• BLMPred demonstrates superior performance compared to existing tools on independent datasets.
• The tool accurately predicts linear B-cell epitopes based solely on sequence information.
• No reliance on 3D protein structures is required for prediction.

Conclusions:

• BLMPred offers a fast, accurate, and accessible method for B-cell epitope prediction.
• The tool can accelerate vaccine design, antibody development, and diagnostic applications.
• Freely available at https://github.com/bdbarnalidas/BLMPred.git, BLMPred supports further research and development.