A Novel Graph Transformer Framework for Predicting Drug-Disease Associations with Structural Awareness

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Accurate drug-disease association (DDA) prediction accelerates therapeutic discovery.
• Graph representation learning models are efficient but limited by structural biases and inability to capture molecular context.

Purpose of the Study:

• To propose SGTL-DDA, a novel graph transformer framework to improve DDA prediction.
• To address limitations in existing models by incorporating structural information and domain-specific knowledge from heterogeneous biological information networks (HBINs).

Main Methods:

• Developed SGTL-DDA, a graph transformer framework integrating meta-path-guided sampling and a multi-level attention mechanism.
• Jointly learned from structural dependencies and attribute semantics in an end-to-end manner.
• Validated performance on two benchmark datasets using a ten-fold cross-validation scheme.

Main Results:

• SGTL-DDA consistently outperformed state-of-the-art methods in Accuracy, F1-score, and AUC.
• Case studies on Alzheimer's disease and breast cancer demonstrated predictive capability.
• Identified known therapeutics and novel repositioning candidates, supported by molecular docking and literature evidence.

Conclusions:

• SGTL-DDA effectively predicts drug-disease associations by leveraging heterogeneous biological information.
• The framework shows promise for identifying novel therapeutic candidates and drug repositioning opportunities.
• SGTL-DDA offers a powerful computational approach for accelerating drug discovery.