LBCE-XGB: A XGBoost Model for Predicting Linear B-Cell Epitopes Based on BERT Embeddings

Area of Science:

• Immunoinformatics
• Computational Biology
• Bioinformatics

Background:

• Accurate detection of linear B-cell epitopes (BCEs) is crucial for vaccine design, diagnostics, and therapeutics.
• Traditional wet-lab methods for BCE identification are costly and time-consuming, failing to meet the demands of large-scale protein data analysis.
• Existing computational methods for BCE prediction show limitations in performance.

Purpose of the Study:

• To develop a novel, accurate, and cost-effective computational method for predicting linear B-cell epitopes.
• To leverage advanced machine learning and natural language processing techniques for improved epitope prediction.

Main Methods:

• Proposed LBCE-XGB, a method utilizing the XGBoost algorithm for linear BCE prediction.
• Employed residue embeddings from a pre-trained, domain-specific BERT model to capture biological sequence information.
• Integrated additional features including amino acid composition and antigenicity scales.
• Determined optimal feature combinations via cross-validation.

Main Results:

• LBCE-XGB achieved a superior AUROC of 0.845 in fivefold cross-validation, outperforming other machine learning and deep learning models.
• On an independent test set, the model attained an AUROC of 0.838, significantly exceeding state-of-the-art methods.
• BERT-derived representations proved effective for linear BCE prediction.

Conclusions:

• LBCE-XGB demonstrates high accuracy and cost-effectiveness in detecting linear B-cell epitopes.
• The study highlights the potential of BERT embeddings as a powerful feature for epitope prediction.
• LBCE-XGB offers a valuable tool for advancing vaccine development and immunodiagnostics.