A spatial hierarchical network learning framework for drug repositioning allowing interpretation from macro to micro...

Area of Science:

• Biomedical network learning
• Computational drug discovery
• Pharmacological modeling

Background:

• Traditional network architectures struggle to link molecular structures with biomedical networks.
• Current methods face challenges in capturing long-range dependencies and complex biological information.

Purpose of the Study:

• To develop a novel framework, SpHN-VDA, for enhanced drug repositioning and virus-drug association identification.
• To model molecular 3D structures and biological associations into a unified network.

Main Methods:

• Introduction of the Spatial Hierarchical Network (SpHN) to model molecular 3D structures and biological associations.
• Development of an end-to-end framework, SpHN-VDA, using triple attention mechanisms.
• Integration of spatial hierarchical information for improved machine understanding of molecular functionality.

Main Results:

• SpHN-VDA outperforms leading models across three datasets, especially in out-of-distribution and cold-start scenarios.
• Demonstrated enhanced robustness against data perturbation (20-40%).
• Identified 25 potential drug candidates and validated predictions via molecular docking with SARS-CoV-2 spike protein.

Conclusions:

• SpHN-VDA effectively enhances drug repositioning and virus-drug association identification.
• The framework accurately identifies critical binding site motifs without protein residue annotations.
• This research shows significant potential for SpHN-VDA in discovering effective treatments for various diseases.