Computational probing protein-protein interactions targeting small molecules

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Interactome studies generate large datasets, revealing diverse protein-protein interactions (PPIs) as potential drug targets.
• Inhibiting PPIs offers improved specificity and broader target search space compared to single protein inhibition.
• Drug target discovery is challenging, necessitating computational methods for efficient candidate identification.

Purpose of the Study:

• To develop a machine learning method for predicting protein-protein interactions (PPIs) as drug targets on a genomic scale.
• To uncover potential associations between drugs and PPIs for accelerated drug target discovery.

Main Methods:

• Constructed a gold-standard dataset of 227 drug-PPI associations involving 63 PPIs and 113 FDA-approved drugs.
• Characterized drugs by chemical structure, ATC codes, and side effects; represented PPI similarity using an S-kernel based on amino acid sequence.
• Utilized Kronecker product kernel for drug-PPI correlation and a support vector machine (SVM) for prediction.

Main Results:

• Validated the PrePPItar method using cross-validation on the gold-standard dataset.
• Confirmed that drug chemical structure, ATC codes, and side-effect information are predictive of PPI targets.
• Demonstrated that integrating multiple data sources enhances PPI target prediction coverage.

Conclusions:

• PrePPItar serves as a valuable tool for identifying protein-protein interaction targets in drug discovery.
• The method provides a general framework for integrating heterogeneous data in drug target discovery.