Learning size-adaptive molecular substructures for explainable drug-drug interaction prediction by substructure-aware...

Area of Science:

• Computational chemistry
• Pharmacology
• Machine learning

Background:

• Drug-drug interactions (DDIs) can lead to unpredictable pharmacological outcomes.
• Understanding the mechanisms behind DDIs is crucial but challenging.
• Current graph neural networks (GNNs) struggle to pinpoint critical substructures for DDI prediction.

Purpose of the Study:

• To develop an advanced GNN model for accurate DDI prediction.
• To enhance the interpretability of DDI prediction by identifying key contributing substructures.
• To address the limitations of existing GNNs in capturing substructure importance.

Main Methods:

• Introduced a substructure-aware graph neural network (SA-DDI).
• Incorporated a novel substructure attention mechanism to capture size- and shape-adaptive substructures.
• Utilized a substructure-substructure interaction module (SSIM) to model interactions between chemical substructures.

Main Results:

• The SA-DDI model outperformed state-of-the-art methods on two real-world DDI datasets.
• Visual interpretation demonstrated the model's sensitivity to drug structural information.
• The model successfully identified key substructures responsible for DDIs.

Conclusions:

• The SA-DDI model significantly improves the generalization and interpretation capabilities of DDI prediction.
• This approach offers a more transparent understanding of the molecular basis of drug-drug interactions.
• SA-DDI represents a significant advancement in computational pharmacology for drug safety assessment.