Title:
A quick and dirty estimate of logP using MPEOE charges and ZAP solvation

Author:
Clayton Springer
Novartis

Abstract:

I have developed a logP model using Charmm-like charges (obtained from
in-house MPEOE model) and PB/SA (implemented by ZAP). Bordner et al
provide 2115 molecules and their measured logP (with an estimated
experimental error of 0.5 logP units). Using the conformations given by
Bordner, MPEOE charges and ZAP default radii, I calculated the ZAP
solvation energy and recorded its components. These are PB solvation,
solvent accessible surface area and direct coulombic. Using the components
given by ZAP as descriptors, I fitted a linear and non-parametric (random
forest) model of logP. The linear model has a rms error of 0.80 logP units
and the random forest model as an rms error of 0.67 logP units. Thus
straightforward application of default radii and linear relationships does
not provide a good model of logP.