MASMDDI: multi-layer adaptive soft-mask graph neural network for drug-drug interaction prediction

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Accurate drug-drug interaction (DDI) prediction is vital for preventing adverse drug reactions in combination therapy.
• Existing DDI prediction methods often neglect crucial chemical substructure interactions, limiting their predictive power.
• There is a need for advanced computational models that capture detailed molecular substructure information for improved DDI prediction.

Purpose of the Study:

• To introduce a novel Multi-layer Adaptive Soft Mask Graph Neural Network (MASMDDI) for enhanced DDI prediction.
• To address the limitations of current methods by incorporating chemical substructure interactions.
• To improve the accuracy and reliability of predicting potential adverse drug events.

Main Methods:

• Developed a multi-layer adaptive soft mask graph neural network to extract relevant substructures from molecular graphs.
• Utilized an attention mechanism to mine substructure features and update latent representations.
• Decomposed drug-drug interactions into pairwise correlations between core drug substructures for optimized feature representation.

Main Results:

• The proposed MASMDDI model significantly outperforms state-of-the-art methods in DDI prediction tasks.
• MASMDDI demonstrates excellent performance in predicting DDIs for previously unknown drugs.
• Achieved high accuracy scores (ACC: 0.9596, AUROC: 0.9903, AUPRC: 0.9894) in a transductive scenario using the DrugBank dataset, exceeding baseline performance by 2%.

Conclusions:

• MASMDDI offers a superior approach to DDI prediction by effectively leveraging chemical substructure information.
• The model's ability to predict interactions for unknown drugs highlights its potential for real-world clinical applications.
• This novel method represents a significant advancement in computational drug safety and discovery.