KLSD: A Curated Kinase-Ligand Database Mapping Selectivity Landscapes and Polypharmacology.

Area of Science:

• Medicinal Chemistry
• Computational Biology
• Pharmacology

Background:

• Kinase inhibitors are crucial in drug discovery, but understanding their selectivity and polypharmacology is complex.
• Existing resources often lack comprehensive quantitative activity data and selectivity profiles.
• Developing predictive models for kinase inhibitor activity and selectivity is essential.

Purpose of the Study:

• To create KLSD, a large-scale curated database of small-molecule kinase inhibitors and their quantitative activity records.
• To develop a dual-task ensemble model for simultaneously predicting kinase inhibitor potency (pAct) and selectivity.
• To provide a valuable resource for advancing kinase inhibitor research and drug discovery.

Main Methods:

• Curated a database (KLSD) of 787,213 small-molecule kinase inhibitors with 1.8 million quantitative activity records across 428 human kinases.
• Developed a dual-task ensemble model using a multibranch residual multilayer perceptron (MLP) augmented with various machine learning and graph network approaches (SVM, RF, XGBoost, CNN, GCN, GAT, RGCN, VAE).
• Utilized continuous potency labels instead of categorical classes for improved predictive resolution.

Main Results:

• The KLSD database provides extensive data on kinase inhibitor selectivity and polypharmacology.
• The dual-task ensemble model achieved high classification accuracies (≥0.84 per kinase, 0.98 overall) when benchmarked on the JAK family (JAK1/2/3, TYK2).
• The model demonstrated strong generalizability across different kinases.

Conclusions:

• KLSD is a comprehensive resource for kinase inhibitor research, facilitating studies on selectivity and polypharmacology.
• The developed ensemble model accurately predicts kinase inhibitor potency and selectivity, offering a powerful tool for drug discovery.
• Both the KLSD database and the predictive models are freely available to the research community.