Drug repurposing with network reinforcement.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning

Background:

• Drug discovery faces low success rates, motivating drug repurposing.
• In silico methods are increasingly used to reduce costs and time.
• Abundant biomedical data facilitates computational approaches.

Purpose of the Study:

• To propose a network-based machine learning algorithm for drug repurposing.
• To demonstrate a framework for constructing and strengthening drug networks using diverse data.
• To enhance the reliability of in silico drug repurposing.

Main Methods:

• Construct a drug network using drug-target protein information.
• Reinforce the network with drug-drug interaction data from literature.
• Employ graph-based semi-supervised learning for drug scoring and recommendation prioritization.

Main Results:

• The enhanced drug network shows increased coverage (4738 to 5442 drugs) and connections (808,752 to 982,361 associations).
• Drug recommendation reliability improved, with an Area Under the Curve (AUC) of 0.89 (up from 0.79).
• Identified 11 candidate drugs for vascular dementia, including amantadine and conotoxin GV.

Conclusions:

• The network-based approach significantly enhances drug network coverage and connectivity.
• The improved drug network leads to more reliable drug repurposing recommendations.
• This method offers a promising strategy for efficient and effective drug repurposing.