Classification Models for Predicting Cytochrome P450 Enzyme-Substrate Selectivity.

Area of Science:

• Pharmacogenomics
• Computational Chemistry
• Drug Metabolism

Background:

• Cytochrome P450 (CYP) enzymes are crucial for drug metabolism.
• Understanding CYP-substrate specificity is vital for drug development and predicting drug interactions.
• Existing methods for CYP substrate prediction can be limited in scope and scalability.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting CYP enzyme substrate specificity.
• To classify drug substrates for specific CYP enzymes (1A2, 2C9, 2D6, 3A4, and others).
• To predict multi-substrate specificities for CYP enzymes.

Main Methods:

• Utilized three machine learning approaches: decision trees, single-label multi-class, and multi-label multi-class models.
• Employed structural and physicochemical properties of drug substrates for prediction.
• Incorporated genetic algorithms for automated and unbiased descriptor selection.

Main Results:

• Achieved >78% accuracy in classifying substrates for four CYP enzymes (1A2, 2C9, 2D6, 3A4).
• Reached >90% accuracy for single-label classification across eight CYP enzymes.
• Attained >80% accuracy for multi-label classification, identifying substrates metabolized by multiple CYPs.

Conclusions:

• Machine learning models based on structural and physicochemical properties can accurately predict CYP enzyme substrate specificity.
• Automated descriptor selection enhances model applicability to larger datasets and more CYP enzymes.
• These predictive models offer a valuable tool for drug discovery and development by elucidating drug metabolism pathways.