Estimation of partition coefficients of organic compounds: local database modeling with uniform-length structure descriptors

Abstract

Octanol-water partition coefficients are predicted by building local models in a database. The structure space is spanned by uniform-length molecular descriptors derived from connectivity or three-dimensional structure, and from the occurrence of selected substructures. Individual models are derived for each structure cluster. In this contribution, various structure representations, clustering algorithms and mathematical models are investigated and the most appropriate procedure is selected.