A general hypergraph learning algorithm for drug multi-task predictions in micro-to-macro biomedical networks.

Area of Science:

• Computational biology
• Drug discovery
• Machine learning

Background:

• Current biomedical networks lack chemical structure and high-order relation integration.
• Drug properties are heavily influenced by chemical substructures.
• Accelerating drug discovery requires advanced network analysis and deep learning.

Purpose of the Study:

• To develop a general hypergraph learning framework for drug-centric heterogeneous networks.
• To introduce drug-substructure relationships into molecular interaction networks.
• To create a multi-branches HyperGraph learning model (HGDrug) for drug multi-task predictions.

Main Methods:

• Constructed a drug-centric heterogeneous network (DSMN) by incorporating drug-substructure relationships.
• Developed a multi-branches HyperGraph learning model named HGDrug.
• Evaluated HGDrug on four benchmark tasks: drug-drug, drug-target, drug-disease, and drug-side-effect interactions.

Main Results:

• HGDrug achieved highly accurate and robust predictions across all four benchmark tasks.
• Outperformed 8 state-of-the-art task-specific models and 6 general-purpose conventional models.
• Demonstrated HGDrug's ability to capture relationships between drugs with similar functional groups.

Conclusions:

• The proposed HGDrug framework effectively integrates drug-substructure information for enhanced drug discovery.
• The drug-substructure interaction networks improve the performance of existing network models.
• HGDrug offers a promising approach for accelerating drug discovery through advanced deep learning on complex biological networks.