Predictive models for cytochrome p450 isozymes based on quantitative high throughput screening data.

Area of Science:

• Biochemistry and Pharmacology
• Computational Chemistry and Cheminformatics
• Drug Metabolism and Toxicology

Background:

• Cytochrome P450 (CYP450) isozymes are critical for metabolizing endogenous and exogenous compounds, including drugs and toxins.
• Unpredictable interactions with CYP450 can compromise drug efficacy and safety, necessitating accurate prediction methods.
• Assessing metabolic stability and potential toxicity of small molecules requires understanding their CYP450 interactions.

Purpose of the Study:

• To develop predictive models for interactions between small molecules and five major human CYP450 isozymes (1A2, 2C9, 2C19, 2D6, and 3A4).
• To utilize quantitative high throughput screening (qHTS) data to train and validate machine learning models.
• To identify key molecular features influencing CYP450 activity and potential toxicity.

Main Methods:

• Support Vector Classification (SVC) models were developed using customized generic atom types.
• A large dataset of over 17,000 compounds screened against five CYP450 isozymes via qHTS was used.
• Data were randomly split into training and testing sets for model optimization and validation.

Main Results:

• Optimized SVC models demonstrated high predictive accuracy for all five CYP450 isozymes, with Area Under the ROC Curve (AUC) values ranging from 0.85 to 0.93.
• Identified important atom types and features align with known CYP450 substrate preferences.
• Novel features with significant discriminatory power for CYP450 actives versus inactives were discovered.

Conclusions:

• The developed SVC models provide a robust computational tool for predicting CYP450 interactions.
• These models can aid in prioritizing drug candidates during discovery and identifying potential environmental chemical toxicity.
• The findings contribute to a better understanding of structure-activity relationships in CYP450-ligand interactions.