Drug-Drug Interaction Predicting by Neural Network Using Integrated Similarity

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Drug-drug interaction (DDI) prediction is crucial for safe drug development and clinical practice.
• Accurate prediction of unknown DDIs remains a significant computational challenge.
• Existing methods often struggle with precision and comprehensive data integration.

Purpose of the Study:

• To propose and evaluate NDD (Neural network-based method for drug-drug interaction prediction), a novel computational approach for predicting unknown DDIs.
• To leverage multiple drug similarity measures for enhanced DDI prediction accuracy.
• To introduce a new neural network architecture for integrating heterogeneous drug data.

Main Methods:

• NDD calculates drug similarities based on substructure, target, side effect, pathway, transporter, and indication data.
• It employs a heuristic similarity selection process followed by nonlinear similarity fusion to create high-level features.
• A neural network is utilized for the final interaction prediction, combined with a novel architecture.

Main Results:

• NDD demonstrated superior performance compared to six machine learning and six graph-based methods on three benchmark datasets.
• Achieved high cross-validation metrics: AUPR (0.830–0.947), AUC (0.954–0.994), and F-measure (0.772–0.902).
• Case studies confirmed NDD's ability to predict numerous unknown drug pairs effectively.

Conclusions:

• NDD is an efficient and accurate method for predicting unknown drug-drug interactions.
• The integration of diverse drug similarities and a novel neural network architecture contributes to its high performance.
• The developed method and data are publicly available to facilitate further research in DDI prediction.