Neural Networks to Calculate CDK2 Inhibition.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning in structural biology

Background:

• Cyclin-dependent kinase 2 (CDK2) is crucial for cell-cycle progression and a key target for anticancer drug development.
• High-resolution structural and binding affinity data for CDK2 enable the creation of predictive computational models.

Purpose of the Study:

• To develop and validate a machine learning workflow for predicting CDK2 inhibition.
• To integrate molecular docking results with experimental binding data for enhanced model accuracy.

Main Methods:

• Utilized Molegro Virtual Docker (MVD) for docking simulations and Molegro Data Modeller for data integration.
• Employed Jupyter Notebooks to connect MVD, data modeling, and machine learning model training.
• Trained neural network models using binding affinity data from BindingDB.

Main Results:

• The integrated workflow generated a neural network model that accurately predicts CDK2 inhibition.
• The developed model demonstrated superior predictive performance compared to traditional scoring functions.
• The workflow successfully integrated docking results with experimental binding data.

Conclusions:

• The presented workflow offers a robust method for predicting kinase inhibition, applicable to any target with known structural and binding data.
• This approach enhances the accuracy of drug-target interaction predictions by combining computational and experimental data.
• The flexible workflow accommodates various structural data types, including experimental and predicted models (e.g., AlphaFold).