The Development and Application of KinomePro-DL: A Deep Learning Based Online Small Molecule Kinome Selectivity...

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Characterizing kinase inhibitor selectivity is crucial for drug discovery, identifying off-target effects, and enabling drug repurposing.
• Experimental kinome selectivity profiling is resource-intensive and time-consuming.

Purpose of the Study:

• To develop a deep learning model for predicting kinome selectivity profiles of small molecules.
• To create a user-friendly web server for predicting kinase inhibitor polypharmacology.

Main Methods:

• A multitask deep neural network was trained on a curated dataset of inhibitors against 191 kinases.
• The model was built by integrating and cleaning data from six public datasets.
• Performance was evaluated using auROC, prc-AUC, Accuracy, and Binary Cross-entropy metrics.

Main Results:

• The model achieved high predictive performance (auROC=0.95, prc-AUC=0.92, Accuracy=0.90).
• It demonstrated strong performance in a priori testing on diverse protein targets.
• Novel CDK2 kinase inhibitors with potent activity and selectivity were identified using the model in a virtual screening workflow.

Conclusions:

• The developed deep learning model accurately predicts kinome selectivity profiles and polypharmacology effects.
• The KinomePro-DL web server provides a valuable tool for researchers to predict inhibitor profiles and fine-tune models.
• This approach accelerates the discovery of safer and more effective kinase inhibitors.