AbCPE

AbCPE predicts the antibody class(es) (IgG, IgE, IgA, and IgM) that sequential B-cell epitopes can bind using multi-label classification to support epitope selection for vaccine and therapeutic antibody development.

Key Features:

• Multi-Label Classification: Predicts which antibody classes (IgG, IgE, IgA, and IgM) an epitope can potentially bind, reflecting that multiple classes may be involved in immune responses.
• Machine Learning Algorithms: Implements Binary Relevance, Label Powerset, Random Forest, and AdaBoost to derive predictive models from epitope data.
• Performance Metrics: Evaluated using Hamming Loss, Precision, Recall, and F1 score, with Binary Relevance achieving Hamming Loss 0.1121 on the training set and 0.1074 on the test set.
• Sequence Features: Uses dipeptide composition as the feature set to represent sequential B-cell epitopes for model training.
• Novelty: Constitutes a multi-label method for predicting antibody class(es) for sequential B-cell epitopes.

Scientific Applications:

• Vaccine Development: Enables prediction of antibody class-specific epitopes to inform the design of vaccines that elicit desired immune responses.
• Therapeutic Antibody Design: Identifies potential epitope targets across multiple antibody classes to support therapeutic antibody development.
• Immunotherapeutics: Informs synthetic biology and immunotherapy research by indicating which antibody classes an epitope can bind.

Methodology:

Algorithms (Binary Relevance, Label Powerset, Random Forest, AdaBoost) were trained on a curated dataset of epitopes with known binding affinities using dipeptide composition as input features.