Gradient Descent to Predict Enzyme Inhibition.

Area of Science:

• Computational chemistry and drug discovery.
• Application of artificial intelligence in bioinformatics.
• Machine learning for molecular modeling.

Background:

• Protein targets are crucial in drug discovery and can be analyzed using machine learning.
• Gradient Descent is a powerful optimization algorithm for machine learning models.
• Predicting enzyme inhibition is key for developing targeted therapies.

Purpose of the Study:

• To describe and apply Gradient Descent methods for predicting protein target inhibition.
• To build a regression model for identifying potential anticancer drugs.
• To demonstrate the integration of machine learning tools in drug discovery pipelines.

Main Methods:

• Utilized Batch Gradient Descent and Stochastic Gradient Descent (SGDRegressor from Scikit-Learn).
• Integrated AutoDock Vina for calculating protein-ligand interaction data.
• Employed the SAnDReS 2.0 program for implementing SGDRegressor models.
• Developed a hands-on approach using Jupyter Notebooks and available datasets.

Main Results:

• Successfully created regression models to predict enzyme inhibition.
• Demonstrated the prediction of inhibition for cyclin-dependent kinase 2, a target for anticancer drugs.
• Combined docking data with machine learning for accurate predictions.

Conclusions:

• Gradient Descent variants, particularly SGDRegressor, are effective for predicting protein target inhibition.
• The developed methodology facilitates the drug discovery process by enabling efficient screening of potential drug candidates.
• Open-source tools and data are provided to support further research in computational drug discovery.