Machine Learning Models to Predict Cytochrome P450 2B6 Inhibitors and Substrates

Area of Science:

• Pharmacogenomics
• Computational Chemistry
• Drug Metabolism

Background:

• Cytochrome P450 2B6 (CYP2B6) metabolizes approximately 7% of marketed drugs.
• Regulatory guidelines (FDA) mandate evaluation of drug interactions with major drug-metabolizing enzymes like CYP2B6.
• Accurate prediction of CYP2B6 inhibitors and substrates is crucial for efficient drug development.

Purpose of the Study:

• To develop and validate machine learning models for predicting CYP2B6 inhibitors and substrates.
• To identify key structural features associated with CYP2B6 inhibition and substrate activity.
• To establish the applicability domain for the developed predictive models.

Main Methods:

• Development of conventional machine learning and deep learning models.
• Evaluation of model performance using 10-fold cross-validation, test sets, and external validation sets.
• Substructural fragment analysis and information gain for feature identification.
• Nonparametric methods for defining the model applicability domain.

Main Results:

• The best CYP2B6 inhibitor model achieved AUC values of 0.95 (cross-validation) and 0.75 (test set).
• The best CYP2B6 substrate model achieved AUC values of 0.93 (cross-validation) and 0.90 (test set).
• Significant substructural fragments influencing CYP2B6 interactions were identified.

Conclusions:

• Developed machine learning models demonstrate strong predictive performance for CYP2B6 inhibitors and substrates.
• The identified substructural fragments provide insights into CYP2B6-drug interactions.
• These models and findings can facilitate early-stage drug discovery by predicting potential CYP2B6 liabilities.