MKMGCN-DDI: Predicting Drug-Drug Interactions via Magnetic Graph Convolutional Network With Multiple Kernels

Area of Science:

• Pharmacology and Cheminformatics
• Artificial Intelligence in Medicine
• Network Science

Background:

• Polypharmacy is prevalent, but predicting drug-drug interactions (DDIs) is challenging due to cost and clinical limitations.
• Existing graph neural network (GNN) approaches for DDI prediction often neglect the multifaceted nature of interactions, including varying effects and asymmetric drug roles.
• A comprehensive understanding of DDIs is crucial for patient safety and effective treatment.

Purpose of the Study:

• To develop advanced graph neural network (GNN) models for predicting comprehensive drug-drug interactions (DDIs).
• To address limitations in current DDI prediction by incorporating interaction types (enhancive/depressive) and asymmetric drug roles.
• To establish new prediction tasks that encompass joint prediction of DDI types and directions.

Main Methods:

• Formal definition of new joint prediction tasks for DDIs, encompassing types and directions.
• Proposal of a unified framework, MKMGCN-DDI, utilizing Magnetic Laplacian matrices and multiple-kernel graph convolutional networks (MKMGCN).
• Implementation of multiple graph filters to encode comprehensive DDI information within the spectral GNN framework.

Main Results:

• The proposed MKMGCN-DDI framework demonstrates strong adaptability across multiple DDI prediction tasks.
• Significant improvements in prediction accuracy were observed compared to existing methods, even on simpler DDI prediction tasks.
• Experimental validation on three datasets confirmed the effectiveness and robustness of the developed approach.

Conclusions:

• The MKMGCN-DDI framework offers a more comprehensive approach to DDI prediction by considering interaction nuances.
• The model's feasibility is supported by case studies on breast and lung neoplasms, with over 50% of top predictions validated.
• This research advances the application of GNNs in predicting complex pharmacological interactions, aiding clinical decision-making.