Trainable structure-activity relationship model for virtual screening of CYP3A4 inhibition

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Pharmacology

Background:

• Cytochrome P450 3A4 (CYP3A4) is a key enzyme in drug metabolism.
• Predicting CYP3A4 inhibition is crucial for drug development and avoiding drug-drug interactions.
• Existing models often have limited applicability domains.

Purpose of the Study:

• To develop a novel structure-activity relationship (SAR) model for predicting CYP3A4 inhibition.
• To enhance the model's applicability domain and accuracy using the GALAS methodology.
• To assess the model's utility for virtual screening of potential drug candidates.

Main Methods:

• Development of a quantitative structure-activity relationship (QSAR) model using data from >800 compounds.
• Application of the Global, Adjusted Locally According to Similarity (GALAS) methodology, combining global QSAR with local similarity corrections.
• Testing and adaptation of the model on PubChem screening data, including high-throughput screening data.

Main Results:

• The GALAS model achieved an overall accuracy of 89% for compounds within its applicability domain.
• The model demonstrated successful adaptation to different IC50 thresholds and novel chemical scaffolds.
• The initial model's applicability was less than 50% of the PubChem database, but expansion was investigated.

Conclusions:

• The GALAS model provides a reliable tool for predicting CYP3A4 inhibition.
• The methodology allows for expanding the model's applicability domain and adapting it to new datasets.
• This approach can aid in early-stage virtual screening to identify potential drug-drug interactions.