How good are publicly available web services that predict bioactivity profiles for drug repurposing?

Area of Science:

• Pharmacology and Cheminformatics
• Computational Drug Discovery

Background:

• Drug repurposing offers an economical pathway to identify new therapeutics.
• Computational tools predict drug bioactivity, revealing hidden pharmacological potential.
• Existing tools utilize chemical similarity or machine learning for target prediction.

Purpose of the Study:

• To evaluate and compare the performance of computational drug repurposing tools.
• To assess tools based on chemical similarity versus machine learning approaches.
• To determine the efficacy of these tools for both known and novel drug indications.

Main Methods:

• Developed two evaluation datasets: 50 known repositioned drugs and 12 newly patented drugs.
• Assessed tool performance using sensitivity metrics for initial and repurposed indications.
• Compared tools employing chemical similarity (e.g., SEA, SwissTargetPrediction) against machine learning methods (e.g., PASS).

Main Results:

• Machine learning methods, particularly PASS Online, demonstrated high sensitivity (up to 1.00) across both datasets.
• Chemical similarity methods showed variable performance, with lower sensitivity for novel indications (e.g., SuperPred at 0.00).
• PASS Online consistently achieved superior sensitivity, especially for identifying novel therapeutic uses of existing drugs.

Conclusions:

• Machine learning-based computational tools are more effective for drug repurposing than chemical similarity methods.
• These findings highlight the advantage of machine learning in predicting novel indications for drug repurposing.
• The study validates PASS Online as a highly sensitive tool for exploring drug repurposing opportunities.