DELPHI

DELPHI predicts protein–protein interaction (PPI) binding sites from protein sequences using deep learning with an ensemble architecture and data augmentation.

Key Features:

• Ensemble architecture: Uses an ensemble structure to combine multiple predictive models for PPI binding site identification.
• Data augmentation: Applies data augmentation techniques to enhance training data diversity and model generalization.
• Feature integration: Leverages both novel and existing sequence-derived features for prediction.
• Deep learning algorithms: Employs advanced deep learning methods for sequence-based PPI binding site prediction.
• Benchmark evaluation: Was evaluated against nine state-of-the-art programs across five datasets demonstrating superior accuracy.
• Reproducible artifacts: Includes a trained model and source code for training, prediction, and data processing.

Scientific Applications:

• PPI binding site prediction: Predicts residue-level protein–protein interaction binding sites from protein sequences.
• Benchmarking of predictors: Enables comparative evaluation of PPI prediction methods against state-of-the-art programs and datasets.
• Study of protein interactions: Supports investigation of protein–protein interactions underlying biological functions.
• Model development and validation: Provides resources for training and validating predictive models for PPI site identification.

Methodology:

Uses an ensemble of deep learning models enhanced by data augmentation, integrates novel and existing sequence-derived features, performs model training, prediction, and data processing, and benchmarks performance against nine state-of-the-art programs on five datasets.