I. Abstract

The partition coefficients for neutral species of 18 quinazolones and related compounds, as calculated by the ACD/Labs algorithm, are shown to have a larger than usual deviation from experimental values. Modification of the logP prediction algorithm, to permit "system training" on two representative species, leads to a four-fold improvement in the accuracy of prediction. The system training method is extended to logD predictions, and is shown to improve the accuracy of predicting a set of 19 phenylpiperidine derivatives two-fold and a set of 10 phenylquinolines three-fold.