A Network Enhancement Method to Identify Spurious Drug-Drug Interactions.

Area of Science:

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

Background:

• Accurate drug-drug interaction (DDI) detection is crucial for medical safety and drug regulation.
• Current graph neural network (GNN) models for DDI prediction are hindered by spurious links in DDI networks.
• Data errors and incorrect drug information can compromise the accuracy of GNN-based DDI predictions.

Purpose of the Study:

• To propose ANSM, a network-enhancement method to identify and attenuate spurious links in DDI networks.
• To improve the accuracy of GNN-based DDI prediction by refining the underlying DDI network structure.
• To address the challenge of data inaccuracies affecting the reliability of computational DDI prediction.

Main Methods:

• ANSM integrates a feature extractor for local structural features, a network optimizer for refining features using network information, and a discriminative classifier to identify spurious links.
• The feature extractor captures relationships between drug node pairs.
• The network optimizer enhances feature extraction and mitigates the impact of false DDI links, followed by classification of potential spurious links.

Main Results:

• Experimental results show that ANSM effectively identifies spurious drug-drug interactions.
• The proposed method demonstrates superior performance compared to existing state-of-the-art approaches in spurious DDI detection.
• ANSM successfully enhances the accuracy and reliability of DDI networks.

Conclusions:

• ANSM provides a robust solution for purifying DDI networks, thereby improving the accuracy of computational DDI prediction.
• By addressing spurious links, ANSM contributes to more reliable GNN-based DDI prediction models for enhanced patient safety.
• The method offers a significant advancement in ensuring the integrity of DDI data used in pharmaceutical research and clinical practice.