DEEPCYPs: A deep learning platform for enhanced cytochrome P450 activity prediction.

Area of Science:

• Pharmacology and Drug Metabolism
• Computational Chemistry and Cheminformatics
• Artificial Intelligence in Drug Discovery

Background:

• Cytochrome P450 (CYP) enzymes are crucial for metabolizing drugs and xenobiotics.
• CYP-mediated drug-drug interactions are a major cause of drug development failure and market withdrawal.
• Five key CYP isoforms (1A2, 2C9, 2C19, 2D6, 3A4) metabolize most approved drugs.

Purpose of the Study:

• To develop in silico classification models for predicting molecular inhibitory activity against five major CYP isoforms.
• To leverage a novel FP-GNN deep learning method for enhanced predictive accuracy.
• To create a user-friendly webserver (DEEPCYPs) for predicting CYP inhibition.

Main Methods:

• Development and application of a multi-task FP-GNN deep learning model.
• In silico classification for predicting inhibitory activity against CYP1A2, 2C9, 2C19, 2D6, and 3A4.
• Rigorous evaluation using metrics like AUC, F1, BA, and MCC, including Y-scrambling tests for validation.

Main Results:

• The multi-task FP-GNN model achieved state-of-the-art predictive performance, with average AUC of 0.905, F1 of 0.779, BA of 0.819, and MCC of 0.647.
• Y-scrambling tests confirmed the model's results were not due to chance.
• The model's interpretability identified critical structural fragments linked to CYP inhibition.

Conclusions:

• The FP-GNN deep learning approach provides highly accurate predictions of CYP inhibition.
• The developed DEEPCYPs webserver can effectively screen compounds for potential CYP inhibitory activity.
• This tool can accelerate drug discovery by identifying unsuitable compounds early and discovering novel CYP inhibitors.