Multi-Label Classification With Dual Tail-Node Augmentation for Drug Repositioning.

Area of Science:

• Computational biology
• Pharmacology
• Machine learning

Background:

• Drug discovery is lengthy and expensive, driving interest in drug repositioning.
• Current machine learning methods for drug repositioning face challenges with insufficient training data and neglect of tail nodes.

Purpose of the Study:

• To propose a novel multi-label classification model for drug repositioning.
• To address limitations in existing methods by incorporating dual tail-node augmentation.

Main Methods:

• Developed Tail-Node Augmentation for Drug Repositioning (TNA-DR) model.
• Integrated k-nearest neighbor (kNN) and contrastive augmentation modules using drug-drug and disease-disease similarity.
• Filtered nodes by degree to focus augmentation on tail nodes.

Main Results:

• Achieved state-of-the-art performance on four real-world datasets.
• Demonstrated effectiveness in identifying drug candidates for new diseases.
• Showcased ability to discover novel drug-disease associations.

Conclusions:

• TNA-DR effectively complements weak supervision in drug-disease associations.
• The model enhances drug repositioning by prioritizing tail nodes.
• TNA-DR offers a promising approach for efficient drug repositioning and discovery.