MINN-DTI

MINN-DTI predicts drug-target interactions by modeling mutual impacts between molecular graphs and protein target distance maps to improve DTI prediction accuracy.

Key Features:

• Joint representation of drugs and targets: Represents drugs as molecular graphs and targets as protein distance maps to capture structural and spatial information.
• Mutual interaction modeling: Uses an interacting-transformer module (Interformer) and an improved Communicative Message Passing Neural Network (Inter-CMPNN) to model bidirectional influences between drug and target representations.
• Enhanced performance: Validated on DUD-E, human, and BindingDB benchmark datasets and reported to outperform existing DTI prediction methods.
• Interpretability: Assigns larger weights to specific amino acids and atoms to indicate contributors to predicted drug-target interactions.

Scientific Applications:

• Drug discovery and lead identification: Predicts likely drug-target interactions to aid identification of potential therapeutic compounds.
• Mechanistic interpretation: Highlights amino acids and atoms contributing to interactions to support elucidation of binding mechanisms.
• Optimization of efficacy and specificity: Provides interaction insights that can inform optimization to improve efficacy and reduce off-target effects.

Methodology:

The Interformer processes molecular graphs and protein distance maps to capture mutual impacts, and the Inter-CMPNN (improved Communicative Message Passing Neural Network) implements communicative message passing between drug and target representations.