kinasepkipred

kinasepkipred predicts ligand-kinase inhibitor constant (pKi) values to estimate binding affinities between ligands and kinases for kinase-targeted drug discovery.

Key Features:

• Machine Learning Models: Employs Random Forest (RFR), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN) models, with RFR selected for superior performance.
• Data Sources: Trained on Ki data compiled from Drug Target Commons (DTC) and Metz databases.
• Feature Utilization: Uses structural and physicochemical features of protein targets and topological pharmacophore atomic triplets fingerprints of ligands.
• Performance Metrics: Internal validation of the selected RFR model reported Pearson R = 0.887, RMSE = 0.475, concordance index = 0.854, and AUC-ROC = 0.957.

Scientific Applications:

• Drug discovery prioritization: Predicts pKi values to prioritize kinase-ligand interactions for selection of potential inhibitors.
• Lead identification and optimization: Supports identification and optimization of candidate kinase inhibitors based on predicted binding affinity metrics.

Methodology:

Collected Ki data from Drug Target Commons (DTC) and Metz; built machine learning models (RFR, XGBoost, ANN) using structural and physicochemical protein features and topological pharmacophore atomic triplets ligand fingerprints; evaluated models using cross-validation and internal validation metrics.