A QSAR investigation of dermal and respiratory chemical sensitizers based on computational chemistry properties

Abstract

A wide range of physicochemical properties based on molecular topology, size and shape, and semi-empirical molecular orbital theory were calculated for a variety of dermal and respiratory sensitizers, as well as some non-active substances. Compounds were randomly selected to belong to a training set of substances (approximately 90%) for development of quantitative structure–activity relationship (QSAR) models or to a test set (approximately 10%) for testing the models. A choice was made of those descriptors which were related to sensitization using standard statistics. Pattern recognition methods were then utilized to identify the combination of properties that provided the greatest contribution to the observed biological effect. Principal components (PC) analysis was then performed on the most important properties. The models derived were then applied to a test set of known sensitizers to predict their class. For dermal and respiratory sensitizers respectively, the PC model classified five (100%) of the R-43 active and two (100%) of the R42-active test set compounds correctly. Analysis of the PC loadings showed that the most useful properties distinguishing respiratory and/or dermal sensitizers from inactive substances were the molecular orbital-based terms.