An improved method for predicting logP

The logarithm of the octanol/water partition coefficient (logP) is used extensively as an indicator of hydrophobicity or lipophilicity within quantitative structure-activity relationships (QSARs) in the pharmaceutical, environmental, and other chemistry-related sciences. LogP is an important component of Lipinki’s “rule of 5” which is used throughout pharmaceutical industry to distinguish “drug-like” compounds from those likely to exhibit poor bioavailability. Using structure-based, calculated predictions of logP, rather than experimentally measured values, is of obvious advantage given the nearinfinite number of potential drug compounds and the relatively small number of compounds for which logP has been measured. MlogP is a simple yet useful structurebased method developed by Moriguchi et al. to estimate logP values. Moriguchi established rules for dividing molecular structures into 13 types of “fragments” and used measured logP values for 1,230 compounds to develop a regression equation based the frequencies with which each fragment occurred in the structure of each compound. Regrettably, Moriguchi’s description of his method is unclear in several respects. Lipinski implemented a program to calculate MlogP values but was forced to make certain assumptions regarding fragment definitions, etc. We will present a revised and definitive algorithm based on Moriguchi’s original concepts but developed using far more robust fragment definitions and statistical methods. The improved algorithm yields significantly better estimates for 6,529 “STARLIST” compounds with experimentally measured logP values.