Extremely Randomized Trees to Determine Binding Affinity.

Area of Science:

• Computational systems biology
• Artificial intelligence in drug discovery

Background:

• Cyclin-dependent kinase 2 (CDK2) is a key target in anticancer drug development.
• Predicting enzyme inhibition is crucial for identifying effective drug candidates.
• Integrating AI with systems biology offers a holistic approach to complex biological systems.

Purpose of the Study:

• To develop robust regression models for predicting enzyme inhibition.
• To specifically predict the inhibition of cyclin-dependent kinase 2 (CDK2) using computational methods.
• To evaluate the performance of Extremely Randomized Trees models in this prediction task.

Main Methods:

• Utilized protein-ligand docking simulations (Molegro Virtual Docker, AutoDock Vina 1.2) to generate data.
• Employed the Extremely Randomized Trees algorithm, implemented in SAnDReS 2.0, for regression modeling.
• Trained and validated models using crystallographic structures and inhibition data for CDK2.

Main Results:

• Extremely Randomized Trees regression models demonstrated superior predictive performance for CDK2 inhibition.
• The developed models outperformed other machine learning techniques evaluated in the study.
• Generated accurate predictions for enzyme inhibition based on docking simulation data.

Conclusions:

• The integration of AI and computational systems biology provides a powerful framework for drug discovery.
• Extremely Randomized Trees models are effective for predicting enzyme inhibition, particularly for targets like CDK2.
• The study provides accessible datasets and code for further research in computational drug design.