Decision Tree for Prediction of Binding Affinity.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Predicting protein-ligand binding affinity is crucial for early-stage drug discovery.
• Machine learning (ML) offers powerful tools for developing accurate scoring functions.
• Integrating docking simulations with ML enhances the modeling of protein systems.

Purpose of the Study:

• To apply Decision Tree regression models for predicting protein-ligand interactions.
• To introduce the SKReg4Model program for exploring scoring function space.
• To provide a practical framework for building predictive models in drug discovery.

Main Methods:

• Utilized Decision Trees, a robust ML technique, for regression analysis.
• Employed features from Vina Force Field and energy terms from docking programs (e.g., Molegro Virtual Docker).
• Developed the SKReg4Model program, based on SAnDReS 2.0 and Scikit-Learn, for model building.

Main Results:

• Demonstrated the efficacy of Decision Trees in modeling complex protein-ligand interactions.
• Showcased the application of SKReg4Model for regression tasks in computational drug discovery.
• Made datasets and a Jupyter Notebook available for reproducibility and further research.

Conclusions:

• Decision Tree regression models are effective for predicting binding affinity.
• SKReg4Model provides a valuable tool for drug discovery research.
• The integration of ML and docking methods advances the field of computational chemistry.