Deep Learning Models for Predicting Human Cytochrome P450 Inhibition and Induction.

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Drug Development

Background:

• Cytochrome P450 enzymes (CYPs) are crucial for drug metabolism, making prediction of their inhibition and induction vital for drug safety.
• Experimental methods for identifying CYP modulators are time-consuming and expensive, necessitating efficient in silico approaches.

Purpose of the Study:

• To develop advanced deep learning models for predicting CYP inhibition and induction.
• To identify structural features associated with CYP modulation.
• To predict human pregnane X receptor (hPXR) activation, a key mechanism for CYP induction.

Main Methods:

• Integration of deep neural networks with principal component analysis (PCA) and synthetic minority oversampling technique (SMOTE) for CYP inhibition prediction.
• Development of a novel classification model for categorizing CYP inhibition strengths (strong, moderate, noninhibitor).
• Application of statistical analysis to identify structural alerts (SAs) for CYP inhibition and induction.
• Creation of a deep learning model for predicting hPXR activation.

Main Results:

• The deep learning model demonstrated excellent predictive performance for CYP inhibition.
• The classification model accurately distinguished between strong, moderate, and noninhibitors of key CYPs.
• Identified specific structural alerts linked to CYP inhibition and CYP3A4 induction.
• The hPXR activation prediction model achieved good performance.

Conclusions:

• Advanced deep learning models offer efficient and accurate in silico prediction of CYP inhibition and induction.
• The identified structural alerts provide valuable insights for drug design and safety assessment.
• Predicting hPXR activation aids in understanding and mitigating CYP induction risks.