Dual-Channel Multiscale Graph Transformer with Adversarial Contrastive Learning and Low-Rank Disentangled Stratified...

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Drug repositioning accelerates the discovery of new therapies.
• Existing computational methods face challenges like representation collapse, noisy data, and poor negative sampling.

Purpose of the Study:

• To introduce MGTAL-DR, a novel graph learning framework to enhance drug repositioning.
• To address limitations of current computational methods in drug discovery.

Main Methods:

• MGTAL-DR utilizes a dual-channel transformer architecture with adversarial contrastive learning.
• It employs a purely negative sampling strategy and parallel graph encoders.
• Methods include diffusion-based propagation for multiscale similarity and meta-path-guided attention for biological semantics.

Main Results:

• MGTAL-DR achieves state-of-the-art performance on three benchmark datasets.
• The framework demonstrates robustness against noise and improves decision boundaries under sparse conditions.
• A case study identified promising therapeutic candidates for Alzheimer's disease.

Conclusions:

• MGTAL-DR offers a significant advancement in computational drug repositioning.
• The framework shows practical utility and real-world potential for identifying novel therapeutics.
• It effectively overcomes key limitations in existing drug discovery methods.