iDrug-Target

iDrug-Target predicts interactions between drug compounds and protein targets to improve interaction discovery for drug design by addressing class imbalance in benchmark datasets.

Key Features:

• Drug–protein interaction prediction: Predicts interactions between drug compounds and protein targets at the compound–target pair level.
• Imbalanced dataset handling: Applies the neighborhood cleaning rule and the synthetic minority over-sampling technique (SMOTE) to balance positive (interactive) and negative (non-interactive) subsets in skewed benchmark datasets.
• Optimized benchmark datasets: Produces optimized benchmark datasets with balanced positive and negative subsets for improved model training and evaluation.
• Specialized sub-predictors: Implements four sub-predictors—iDrug-GPCR for G-protein-coupled receptors (GPCRs), iDrug-Chl for ion channels, iDrug-Ezy for enzymes, and iDrug-NR for nuclear receptors—to target specific protein families.
• Validation: Validates predictions against experiment-confirmed datasets.
• Misclassification reduction: Specifically addresses the tendency to misclassify interactive pairs in datasets with disproportionately more non-interactive pairs.

Scientific Applications:

• Drug–protein interaction discovery: Enables identification of potential interactions between small-molecule drugs and protein targets.
• Family-specific interaction identification: Identifies interactions specific to GPCRs, ion channels, enzymes, and nuclear receptors using dedicated sub-predictors.
• Support for drug design: Provides interaction predictions to inform drug design and target prioritization.
• Benchmark preparation for computational studies: Generates balanced benchmark datasets for training and evaluating predictive models.

Methodology:

Imbalanced benchmark datasets are processed using the neighborhood cleaning rule and SMOTE to create optimized datasets; four specialized sub-predictors (iDrug-GPCR, iDrug-Chl, iDrug-Ezy, iDrug-NR) perform interaction predictions, which are validated against experiment-confirmed datasets.