DiscoTope-3.0

DiscoTope-3.0 predicts B-cell epitopes from solved and predicted protein structures using inverse folding latent representations and a positive-unlabelled learning strategy.

Key Features:

• Structure Representation: Employs inverse folding structure representations for epitope prediction.
• Learning Strategy: Uses a positive-unlabelled learning strategy to train the prediction model.
• Structure Compatibility: Maintains high predictive performance across solved, relaxed, and predicted structures.
• Epitope Types: Accurately predicts both linear and conformational epitopes.
• Large-Scale Prediction: Interfaces with RCSB and AlphaFoldDB to facilitate prediction across over 200 million cataloged proteins.
• Batch Processing: Processes over 100 structures per submission.

Scientific Applications:

• Vaccine Development: Supports computational identification of B-cell epitopes for vaccine design.
• Therapeutic Development: Enables identification of epitopes for therapy development.
• Diagnostic Tool Development: Facilitates epitope identification for diagnostic applications.
• Large-Scale Epitope Screening: Allows prediction across millions of proteins without requiring experimental structures.

Methodology:

DiscoTope-3.0 utilizes inverse folding latent representations combined with a positive-unlabelled learning strategy, adapted for both solved and predicted protein structures.