FAME 3

FAME_3: Prediction of metabolic sites of metabolism (SoMs)

FAME_3 predicts sites of metabolism (SoMs) for phase 1 and phase 2 metabolic reactions using models trained on the MetaQSAR database, which contains over 2100 substrates and more than 6300 experimentally confirmed SoMs, including redox reactions, hydrolysis, nonredox reactions, and conjugation reactions.

Key Features:

• MetaQSAR Training Data: Trained on >2100 substrates with >6300 experimentally validated SoMs derived from MetaQSAR (Pedretti et al., Journal of Medicinal Chemistry, 2018), covering diverse metabolic reaction types.
• Global Metabolism Model: Predicts SoMs across both phase 1 and phase 2 metabolism (MCC = 0.50).
• Phase 2-Specific Model: Predicts SoMs specific to phase 2 metabolism (MCC = 0.75).
• Cytochrome P450 Model: Predicts SoMs mediated by cytochrome P450 enzymes (MCC = 0.57).
• Applicability Domain Estimation (FAMEscore): Implements an atom-based distance metric using nearest-neighbor search in atom environment space to estimate model applicability for individual predictions.

Scientific Applications:

• Drug Metabolism Prediction: Identifies metabolic liabilities in synthetic compounds, natural products, and derivatives to support pharmacokinetic and safety assessment in drug discovery.

Methodology:

Machine learning models trained on atom-centered representations derived from the MetaQSAR dataset predict SoMs for phase 1 and phase 2 reactions. Model reliability is assessed using Matthews correlation coefficient (MCC), and prediction confidence is estimated through FAMEscore, which quantifies atom-level similarity to training data via nearest-neighbor distance in atom environment space.