PrePPItar

PrePPItar predicts drug–protein–protein interaction (drug–PPI) associations by integrating chemical-structure, ATC-code, and side-effect similarities with PPI sequence-derived S-kernel similarity via a Kronecker product kernel and a support vector machine trained on a gold-standard dataset.

Key Features:

• High-throughput interaction analysis: Leverages large-scale interactome datasets that provide genome-scale measurements of protein–protein interactions (PPIs).
• Machine learning approach: Uses a support vector machine trained on a gold-standard dataset of 227 associations linking 63 PPIs and 113 FDA-approved drugs.
• Data integration framework: Integrates drug similarity from chemical structure, Anatomical Therapeutic Chemical (ATC) codes, and side-effect profiles with PPI similarity via a symmetrical S-kernel derived from protein amino-acid sequences.
• Kronecker product kernel: Couples drug and PPI similarity spaces using a Kronecker product kernel to infer candidate drug–PPI associations.
• Validation and predictive power: Evaluates performance by cross-validation on the reference dataset, with chemical structure, ATC codes, and side-effect information each contributing to predictive accuracy.
• Enhanced prediction coverage: Integrates diverse data sources to expand coverage of drug–PPI target prediction and prioritize candidates for experimental validation.

Scientific Applications:

• Drug discovery: Extends target identification from single proteins to protein–protein interactions to enable higher specificity and reduced off-target effects.
• Target prioritization: Proposes novel drug–PPI associations to systematically prioritize candidates for experimental follow-up and therapeutic target exploration.

Methodology:

Manually curate a gold-standard positive set (227 associations linking 63 PPIs and 113 FDA-approved drugs); represent drugs using chemical-structure features, ATC-code annotations, and side-effect similarity; encode PPI similarity with a symmetrical S-kernel derived from protein amino-acid sequences; link drug and PPI representations via a Kronecker product kernel; and train and validate a support vector machine using cross-validation.