In silico prediction of chemical metabolism by human UDP glucuronosyltransferase isoforms: Evaluation of classification algorithms

UDP-glucuronosyltransferase (UGT) is an enzyme ‘‘superfamily’’ involved in the metabolism
of drugs, nondrug xenobiotics, and endogenous compounds. Using simple chemical properties/descriptors, partial least squares discriminant analysis (PLSDA), Bayesian regularized
artificial neural network (BRANN) and support vector machine (SVM) methodologies were
compared for their ability to classify a large number of known substrates and nonsubstrates of
12 human UGT isoforms. In general, the SVM methodology was best able to predict the testset chemicals (30% of the data that was not used in model generation), followed by BRANN
and then PLSDA. The test sets of five out of the twelve isoforms were predicted with greater
than 80% accuracy using the SVM methodology. These results represent the first use of
pattern recognition methods to discriminate between large datasets of diverse substrates
and nonsubstrates of human drug metabolizing enzymes and the first detailed comparison
of PLSDA, BRANN, and SVM performance with metabolic datasets