Prediction of UGT-mediated Metabolism Using the Manually Curated MetaQSAR Database.

Area of Science:

• Computational chemistry
• Drug metabolism
• Pharmacokinetics

Background:

• Glucuronidation is the most frequent metabolic conjugation reaction, crucial for drug clearance.
• Computational approaches for predicting glucuronidation are underdeveloped despite its significance.
• Existing methods lack comprehensive predictive power for UDP-glucuronosyltransferase (UGT) activity.

Purpose of the Study:

• To develop and validate computational models for predicting UGT-mediated metabolism.
• To predict the occurrence of glucuronidation reactions using molecular descriptors.
• To differentiate between O- and N-glucuronidation pathways.

Main Methods:

• Utilized the MetaQSAR metabolic reaction database for model generation.
• Employed the Random Forest algorithm with molecular descriptors.
• Developed two models: one for reaction occurrence prediction and another for O- vs. N-glucuronidation classification.
• Validated models using internal metrics (MCC, AUC) and an external test set.

Main Results:

• The first model achieved high internal validation (MCC=0.76, AUC=0.94) and external validation (MCC=0.70, AUC=0.90) for predicting glucuronidation occurrence.
• The second model successfully distinguished between O- and N-glucuronidations.
• Model optimization via random undersampling significantly improved minority class recall for O-/N-glucuronidation prediction (from 0.55 to 0.78).

Conclusions:

• Developed robust computational models for predicting UGT-mediated metabolism.
• These models offer valuable tools for drug discovery and development by forecasting metabolic fate.
• The models provide accurate predictions for both the likelihood and type of glucuronidation.