A Multi-Label Learning Framework for Drug Repurposing.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repurposing is crucial for discovering new therapeutic uses for existing drugs.
• Current drug-target interaction prediction methods often rely on extensive, experimentally unverified negative data, reducing prediction reliability.
• A need exists for more credible and efficient drug repurposing strategies.

Purpose of the Study:

• To propose a multi-label learning framework for drug repurposing and new drug discovery.
• To model inter-drug associations explicitly using experimentally observed data.
• To reduce data constraints by not requiring drug chemical or protein structures.

Main Methods:

• Treated each drug as a class label and its known target genes as class-specific training data.
• Employed an l2-regularized logistic regression model for supervised learning.
• Utilized stratified multi-label cross-validation and independent testing on DrugBank data.

Main Results:

• Achieved 84.9% accuracy in recognizing known target genes for at least one drug.
• Demonstrated 86.73% accuracy in predicting independent drug-target interactions (DTIs).
• Showcased the framework's ability to generalize across a large drug/target space without structural information.

Conclusions:

• The proposed multi-label learning framework offers a robust and credible approach to drug repurposing and discovery.
• The model successfully predicts novel drug-target associations and identifies new therapeutic potential.
• Findings support potential new clinical therapies by linking drugs to diseases and phenotypes.