A Molecular Fragment Representation Learning Framework for Drug-Drug Interaction Prediction

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery

Background:

• Concurrent drug use poses risks due to potential drug-drug interactions (DDIs).
• Accurate DDI prediction is vital for drug development and patient safety.
• Existing methods often overlook molecular hierarchical information and use restrictive substructure definitions.

Purpose of the Study:

• To develop an advanced computational framework for precise DDI prediction.
• To integrate molecular hierarchical information into DDI prediction models.
• To overcome limitations of existing substructure-based DDI prediction methods.

Main Methods:

• A novel molecular fragment representation learning framework was developed.
• A fragment extraction module was designed to obtain molecular fragments.
• Molecular hierarchical information was integrated to capture comprehensive features for pairwise fragment interaction analysis.

Main Results:

• The proposed method achieved state-of-the-art performance on DrugBank and Twosides datasets.
• Accuracy for predicting interactions involving unseen drugs improved by over 20%.
• Case studies confirmed the identification of crucial, functionally relevant substructures influencing DDIs.

Conclusions:

• The developed framework significantly enhances DDI prediction performance.
• The method offers high interpretability by identifying key interacting substructures.
• This approach represents a significant advancement for computational drug safety assessment.