On the PNN Modeling of Estrogen Receptor Binding Data for Carboxylic Acid Esters and Organochlorine Compounds†

We describe the relationship between the estrogen receptor binding and the molecular structure of chemicals using the probabilistic neural network methodology with structural fragment descriptors as input variables and a data set of 1118 compounds. Exploratory models identified two subsets of chemicals for which the predictions were well correlated with the measured values, namely chlorine-containing compounds and carboxylic esters, and for which individual models were developed. Both compound classes are in the classification system for chemicals on the Canadian Domestic Substances List (DSL) and the data cover five orders of magnitude in activity in each of these classes. The results show excellent performance of both models and are highly encouraging in the search for other models for this and other receptor binding data as well as other classes of DSL substances. They also confirm the flexibility, usefulness and applicability of the probabilistic neural networks as modeling methodology to a wide variety of modeling challenges in the environmental and health fields.