Two-step Multivariate Classification of the Mechanisms of Toxic Action of Phenols

A two-step classification algorithm was developed for the classification and prediction of the mechanisms of toxic action of phenols. The mechanisms considered were polar narcosis, weak acid respiratory uncoupling, pro-electrophilicity, and soft electrophilicity. The classification algorithm was based on linear discriminant analysis using molecular descriptors as discriminating variables. Polar narcosis/pro-electrophilicity was separated from weak acid respiratory uncoupling/soft electrophilicity in the first step. In the second step, polar narcosis was separated from pro-electrophilicity and weak acid respiratory uncoupling was separated from soft electrophilicity. Different molecular descriptors were involved as discriminating variables at different steps of the classification algorithm. This algorithm classified any given phenol as acting by one and only one of the four mechanisms considered. Compared to one-step classification models, the prediction accuracy of the two-step algorithm was higher for all four mechanisms involved.