Development of in silico classification models for binding affinity to the glucocorticoid receptor

Abstract

The endocrine disrupting properties of chemicals acting through the glucocorticoid receptor (GR) have attracted considerable interest. Since there are few data for most chemicals on their endocrine properties in silico approaches seem to be the most appropriate tool for screening and prioritizing chemicals for planning further experiments. In this work, we developed classification models for binding affinity to the glucocorticoid receptor using the counterpropagation artificial neural network method. We considered two series of 142 and 182 compounds and their binding affinity to the glucocorticoid receptor as agonists and antagonists, respectively. The compounds belong to different chemical classes. The compounds were represented by a set of descriptors calculated with the DRAGON program. The clustering structure of sets was studied with standard principal component method. A weak separation between binders and non-binders was found. Another classification model was developed using the counterpropagation artificial neural network method (CPANN). The final classification models developed were well balanced and showed a high level of accuracy, with 85.7% of GR agonist and 78.9% of GR antagonist correctly assigned in leave-one-out cross-validation.

Keywords: Binding to glucocorticoid receptor; Counterpropagation artificial neural networks; DRAGON descriptors; Endocrine disruptions; In silico classification.