SADR: Self-Supervised Graph Learning With Adaptive Denoising for Drug Repositioning.

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Traditional drug development is costly and time-intensive.
• Drug repositioning offers a faster, more cost-effective alternative.
• Graph representation learning methods show promise for drug repositioning but struggle with sparse and noisy data.

Purpose of the Study:

• To propose a robust drug repositioning model that overcomes limitations of existing graph-based methods.
• To enhance prediction accuracy in sparse datasets and improve robustness against noise.

Main Methods:

• Developed SADR (Self-supervised graph learning with Adaptive Denoising) model.
• Employed data augmentation and contrastive learning for node feature representation.
• Integrated an adaptive denoising training (ADT) component to mitigate noise impact.

Main Results:

• SADR demonstrated superior prediction accuracy over baseline models across three datasets.
• The model effectively addresses challenges posed by data sparsity and noise.
• Identified top 10 potential approved drugs for treating two specific diseases.

Conclusions:

• SADR offers a robust and effective approach for drug repositioning.
• The model's ability to handle data imperfections enhances its utility in drug discovery.
• SADR successfully identifies novel drug candidates for disease treatment.