Tree-Based Methods to Predict Enzyme Inhibition.

Area of Science:

• Computational biology
• Cheminformatics
• Machine learning in drug discovery

Background:

• Biological systems, including proteins, are complex and suitable for machine learning modeling.
• Cyclin-dependent kinase 4 (CDK4) is a crucial cell-cycle regulator and a target for anticancer therapies.

Purpose of the Study:

• To develop and evaluate computational regression models for predicting CDK4 inhibition.
• To identify machine learning techniques that offer superior predictive performance for drug discovery targeting CDK4.

Main Methods:

• Utilized tree-based machine learning algorithms: Decision Trees, Extra Trees, and Random Forest.
• Engineered predictive features using Molegro Virtual Docker (MVD).
• Implemented 80 regression methods via the Scikit-Learn library in the SKReg4Model Jupyter Notebook.

Main Results:

• The Extra Trees regression model achieved the highest predictive performance.
• The developed Extra Trees model outperformed the classical Plants scoring function used in docking programs.
• All datasets and code are publicly available on GitHub for reproducibility.

Conclusions:

• Machine learning, specifically Extra Trees, provides a robust approach for modeling and predicting CDK4 inhibition.
• This study highlights the potential of advanced computational methods in accelerating the discovery of novel anticancer drugs targeting CDK4.