Boosting Drug-Disease Association Prediction for Drug Repositioning via Dual-Feature Extraction and Cross-Dual-Domain...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Discovery

Background:

• Drug repositioning offers a cost-effective approach to drug development by identifying new therapeutic uses for existing drugs.
• Current methods often overlook inter-relationships between drug and disease feature spaces, leading to biased association information.
• Accurate encoding of drug and disease features is crucial for effective drug repositioning.

Purpose of the Study:

• To propose a novel neural network model, the Dual-Feature Drug Repurposing Neural Network (DFDRNN), for more accurate drug-disease association prediction.
• To address limitations in existing methods by incorporating dual-feature extraction from drug-disease networks.
• To enhance the encoding of drugs and diseases by considering both similarity and association features.

Main Methods:

• Developed the Dual-Feature Drug Repurposing Neural Network (DFDRNN) model.
• Utilized a self-attention mechanism for neighbor feature extraction.
• Incorporated single-domain (SDDFE) and cross-domain (CDDFE) dual-feature extraction modules.
• Designed a cross-dual-domain decoder for predicting drug-disease associations.

Main Results:

• The DFDRNN model demonstrated superior performance compared to six state-of-the-art methods on four benchmark datasets.
• Achieved an average AUROC of 0.946 and an average AUPR of 0.597.
• Case studies confirmed the model's applicability in real-world drug repositioning scenarios.

Conclusions:

• The proposed DFDRNN model significantly improves the accuracy of drug repositioning by effectively mining dual features from drug-disease networks.
• DFDRNN offers a promising approach for identifying novel indications for existing drugs.
• The model's performance and real-world applicability highlight its potential to accelerate drug discovery and development.