GLORYx

GLORYx predicts and ranks chemical structures of human xenobiotic metabolites resulting from phase 1 and phase 2 biotransformations by combining machine learning site-of-metabolism predictions with reaction-rule-based generation to support drug discovery and metabolism research.

Key Features:

• Phase 1 and Phase 2 coverage: Predicts metabolites formed by both phase 1 and phase 2 biotransformations, extending beyond cytochrome P450-mediated phase 1 reactions.
• Machine learning SoM prediction: Uses machine learning-based site-of-metabolism (SoM) predictions to identify likely metabolic sites.
• FAME 3 models: Leverages FAME 3 machine learning models to generate SoM probability scores.
• Reaction rule sets: Applies comprehensive reaction rule sets to enumerate candidate metabolite structures.
• Scoring and ranking: Uses SoM probabilities to score and rank predicted metabolite structures.
• Performance metrics: Achieves 77% recall and an AUC of 0.79 on a manually curated test dataset containing phase 1 and phase 2 metabolites.
• Data and rules availability: Associated datasets and reaction rules are provided to support further research and application.

Scientific Applications:

• Regioselectivity prediction: Predicts likely sites of metabolic modification to inform regioselectivity analyses.
• Drug discovery and development: Anticipates metabolic fate of xenobiotics to support lead optimization and safety assessment.
• Metabolite prioritization: Ranks candidate metabolite structures to prioritize experimental follow-up.
• Metabolic pathway exploration: Generates plausible metabolite structures for investigation of human metabolic pathways.

Methodology:

Integrates FAME 3 machine learning SoM predictions with comprehensive reaction rule sets to generate candidate metabolites and uses SoM probabilities to score and rank predicted structures; evaluated on a manually curated test dataset reporting recall and AUC.