AutoDDI: Drug-Drug Interaction Prediction With Automated Graph Neural Network.

Area of Science:

• Bioinformatics
• Computational Chemistry
• Pharmacology

Background:

• Drug-drug interactions (DDIs) pose significant health risks, including toxicity and reduced efficacy.
• Molecular structures, represented as graphs, are crucial for understanding DDI mechanisms.
• Current methods rely on handcrafted graph neural network (GNN) models, which are labor-intensive and require expert knowledge.

Purpose of the Study:

• To develop an automated method for designing GNN architectures for DDI prediction.
• To eliminate the need for manual GNN architecture design in DDI prediction.
• To improve the efficiency and accuracy of DDI prediction.

Main Methods:

• Designed a comprehensive search space for GNN architectures relevant to DDI prediction.
• Employed a reinforcement learning search algorithm to automatically discover optimal GNN architectures.
• Validated the proposed method, AutoDDI, on two real-world DDI datasets.

Main Results:

• AutoDDI achieved superior performance compared to existing methods on benchmark datasets.
• The automated approach successfully identified key drug substructures contributing to DDIs.
• Visual interpretation confirmed AutoDDI's ability to capture relevant molecular features.

Conclusions:

• AutoDDI offers an efficient and effective solution for automated GNN architecture design in DDI prediction.
• The method reduces reliance on expert experience, accelerating DDI research.
• AutoDDI holds promise for improving drug safety and efficacy through accurate DDI prediction.