CyProduct

CyProduct predicts metabolic byproducts generated by human cytochrome P450 (CYP450) enzymes to characterize CYP-mediated Phase I metabolism for drug discovery and toxicology applications.

Key Features:

• Modular Architecture: Comprises three integrated modules—CypReact predicts whether a query compound will react with a specific CYP450 enzyme; CypBoM (Cytochrome P450 Bond of Metabolism) identifies the chemical bond site likely to undergo metabolic transformation; MetaboGen generates potential metabolic byproducts from bond-site predictions.
• Bond of Metabolism (BoM) concept: Specifies the chemical bonds modified or formed during CYP-mediated reactions, extending traditional site-of-metabolism predictions by focusing on bonds rather than only atomic sites.
• Training database: Utilizes a dataset of 1845 CYP450-mediated Phase I reactions to train the CypBoM predictor.
• Machine learning models: Employs machine learning techniques for reaction prediction, bond identification, and metabolite generation.
• Performance metrics: Reports cross-validated Jaccard scores of 0.380–0.452 for reactive bond prediction across nine major human CYP450 enzymes and an average ~200% improvement in metabolite prediction accuracy compared to ADMET Predictor, BioTransformer, and GLORY.

Scientific Applications:

• Drug discovery: Predicts CYP450-mediated metabolites to inform lead optimization and metabolite profiling.
• Toxicology: Anticipates metabolic byproducts to support safety assessment and identification of potentially toxic metabolites.
• Safety and dosing optimization: Supports anticipation of adverse effects, optimization of dosing regimens, and design of safer pharmaceutical compounds.

Methodology:

Uses machine learning models where CypReact identifies potential CYP450 reactions, CypBoM predicts metabolically labile bonds trained on 1845 CYP450-mediated Phase I reactions, and MetaboGen generates possible byproducts from those bond-site predictions.