MDeePred

MDeePred predicts receptor–ligand binding affinities using multi-channel deep chemogenomic modeling to support drug discovery and drug repurposing by identifying compound–protein interactions.

Key Features:

• Novel Protein Featurization Approach: Integrates sequence, structural, evolutionary, and physicochemical protein features into multiple 2D vectors for input to deep learning models.
• Deep Learning Integration: Employs pairwise input hybrid deep neural networks to predict real-valued compound–target protein interactions (binding affinities).
• Proteochemometric Approach: Combines compound and target protein features at the input level to model interactions comprehensively.
• Scalability and Versatility: Scalable method with demonstrated high predictive performance across benchmark datasets and adaptable featurization for other protein-related predictive tasks.
• Experimental Validation: Predictions have been validated by in vitro comparative analysis of selected kinase inhibitors on cancer cells.

Scientific Applications:

• Computational Drug Discovery and Repositioning: Predicts binding affinities to identify potential therapeutic candidates and prioritize compounds for testing.
• Off-Target and Mechanism Analysis: Aids in understanding drug mechanisms and potential off-target interactions through predicted compound–protein affinities.
• Feature Transfer for Protein Prediction Tasks: Multi-channel featurization can be adapted for other protein-related predictive modeling tasks.

Methodology:

Integrates sequence, structural, evolutionary, and physicochemical protein features into multiple 2D vectors, combines these with compound features in a proteochemometric input, and processes pairwise inputs using hybrid deep neural networks to predict real-valued binding affinities.