Abstract

Quantitative structure pharmacokinetic relationship (QSPKR) modeling can be used to predict the biliary clearance and percentage of dose eliminated in bile (PD_b) in humans before clinical studies. Recently, a QSPKR model based on in-house compounds was derived using simple physicochemical descriptors to predict the PD_b in rats (Drug Metab Dispos 38:422–430, 2010). Our objective was to evaluate the QSPKR model derived for the prediction of PD_b for our larger dataset of 164 compounds in the rat and for the 97 compounds in our human dataset (AAPS J 11:511–525, 2009). Re-analysis of the published QSPKR model revealed the model to be statistically insignificant (Drug Metab Dispos 38:422–430, 2010). Thus, a new statistically significant QSPKR model, consisting of one less descriptor than the published model, was derived from the published data. The newly derived model performed as well as the published model in predicting the PD_b for the training and test sets (Drug Metab Dispos 38:422–430, 2010). In contrast, the new model performed poorly in predicting the PD_b for our larger rat (r² = 0.253) and human dataset (r² = 0.013). The poor predictions for our datasets may be due, in part, to the quality and diversity of the data used to derive and test the model. Our reevaluation suggests that hepatobiliary excretion is a process that cannot truly be captured by simple physicochemical descriptors when examining chemically dissimilar compounds. A simple approach involving a limited number of physicochemical predictors may be useful when examining a structurally similar series of compounds.