DLBEpitope

DLBEpitope predicts linear B-cell epitopes from peptide sequences using deep learning to support vaccine design, clinical diagnostics, and antibody production.

Key Features:

• Data-Driven Approach: Trained on 240,563 peptide samples from the Immune Epitope Database (IEDB), comprising 25,884 linear B-cell epitopes and 214,679 non-epitopes.
• Deep Learning Methodology: Employs feedforward deep neural networks and adapts peptides to uniform lengths by trimming or extending.
• Ensemble Prediction Model: Uses eleven classifiers in an ensemble and classifies a peptide as an epitope only if all eleven classifiers agree.
• Performance Evaluation: Assesses performance using the area under the ROC curve (AUC), observing that AUC increases with peptide length and stabilizes at length 38.
• Robustness and Validation: Produces reproducible results and outperforms existing major models when tested on internal and two public datasets.

Scientific Applications:

• Vaccine design: Guides selection of linear B-cell epitopes for antigen and vaccine candidate prioritization.
• Clinical diagnostics: Identifies linear B-cell epitopes that can serve as diagnostic markers for serological assays.
• Antibody production: Informs epitope selection to optimize antibody generation and specificity.

Methodology:

Computational methods explicitly include feedforward deep neural networks, peptide length normalization by trimming or extending to uniform lengths, an ensemble of eleven classifiers requiring unanimous agreement for epitope calls, evaluation by ROC AUC (with AUC stabilizing at peptide length 38), and validation on internal and two public datasets.