Bioactivity predictions and virtual screening using machine learning predictive model.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Machine learning

Background:

• Predictive modeling for enzyme inhibitors is crucial but limited by chemical bias and lack of repeatability.
• Existing models often fail to account for chemical diversity, hindering drug discovery efforts.

Purpose of the Study:

• To develop a novel machine learning model for predicting enzyme inhibitors that addresses chemical bias.
• To evaluate the model's efficacy using Dipeptidyl peptidase 4 (DPP-4) inhibitors and validate its performance.

Main Methods:

• Development of a machine learning model utilizing Random Forest algorithm.
• Comparison of model performance using diverse training/test data versus random split.
• In-silico screening of the Drug Bank database for DPP-4 inhibitors.
• Validation through molecular docking and molecular dynamics simulations.

Main Results:

• Random Forest algorithm demonstrated the highest accuracy among tested machine learning algorithms.
• The developed model, based on Murcko scaffolds, effectively addressed chemical bias concerns.
• In-silico screening identified two known DPP-4 inhibitors from the Drug Bank database.
• Molecular docking and dynamics simulations confirmed the model's predictive credibility.

Conclusions:

• Machine learning models incorporating Murcko scaffolds can overcome chemical bias in enzyme inhibitor prediction.
• The developed model shows promise for efficient drug discovery and potential clinical translation.
• The approach offers a reliable method for identifying novel drug candidates.