A Hierarchical QSAR Model for Urinary Excretion of Drugs in Humans as a Predictive Tool for Biotransformation

Of the many pharmacokinetic endpoints applicable to in silico screening, drug biotransformation seen as a hybrid, multi-enzymatic disposition parameter, has been little addressed. The aim of this study was to model drug biotransformation, utilising metabolism data for a heterogeneous group of drugs. The data were the cumulative amount of unchanged drug excreted in the urine, expressed as percent of the intravenous dose, administered for 160 drugs. The data were categorised into classes according to excretion ranges. The cut-off values between those ranges were defined so as to enable optimal modelling. For each drug, a total of 72 physicochemical and structural descriptors were calculated. Modelling of the drug metabolism data was attempted utilising a hierarchical approach comprising a set of rules combining both linear discriminant analysis and recursive partitioning. The model developed into a decision tree involving the following descriptors: LogD 6 . 5 , counts of H-bond donors, ionisation potential, COSMIC total energy, electronic energy, counts of OH-groups and COOH-groups and the sum of the total net charges. Overall, this model assigned 90% of the compounds correctly to the categories of extensively, or non-extensively, metabolised. The model was successfully validated using an external test set of 40 compounds.