We are running on a project to construct virtual e-Liver system, which is mainly focused on drug metabolism modeling of real human liver. In general, drug metabolism in liver is occurred two steps, phase I and phase II, to convert lipophilic substrates into more easily excreted polar products. Phase I transformations introduce or unmask a functional group such as oxygenation and hydrolysis and phase II transformations conjugate a highly polar group. In this poster, we present our study on in silico enzyme selectivity model of UDP-glucuronosyltransferase (UGT) and Sulfotransferase (SULT) substrate. Conjugation reactions of these enzymes share similar functional groups so enzyme selectivity discrimination of a chemical compound is important. We made a multi-class classification model which determines phase II transformation fate; substrate of UGT, substrate of SULT, substrate of both enzymes.
The liver is the primary site of drug metabolism in the body. Typically, phase II metabolic conversion of a drug results in inactivation, detoxification, and excretion in further process. Sulfation, glucuronidation, N-acetylation and glutathione conjugation represent the four most prevalent class of phase II metabolism, which may occur directly on the parent compounds that contain appropriate structural motifs, or, as is usually the case, on functional groups added or exposed by phase I oxidation. These four conjugation reactions increase the molecular weight and water solubility of the compound, in addition to adding a negative charge to conjugates. This study was to explore the phase II metabolic fate in combination with a statistical learning method for predicting selectivity of these phase II enzyme. The methodologies used was Recursive Partitioning, an easy and quick method to implement and Random Forest, an ensemble of recursive partitioning trees created using bootstrap samples of the training dataset and random feature selection in tree induction.