Correspondence factor analysis of steroid libraries

Abstract

The receptor binding of a library of 187 steroids to five steroid hormone receptors (estrogen, progestin, androgen, mineralocorticoid, and glucocorticoid) has been analyzed by correspondence factor analysis (CFA) in order to illustrate how the method could be used to derive structure-activity-relationships from much larger libraries. CFA is a cartographic multivariate technique that provides objective distribution maps of the data after reduction and filtering of redundant information and noise. The key to the analysis of very complex data tables is the formation of barycenters (steroids with one or more common structural fragments) that can be introduced into CFA analyses used as mathematical models. This is possible in CFA because the method uses χ 2-metrics and is based on the distributional equivalence of the rows and columns of the transformed data matrix. We have thus demonstrated, in purely objective statistical terms, the general conclusions on the specificity of various functional and other groups derived from prior analyses by expert intuition and reasoning. A finer analysis was made of a series of A-ring phenols showing the high degree of glucocorticoid receptor and progesterone receptor binding that can be generated by certain C-11-substitutions despite the presence of the phenolic A-ring characteristic of estrogen receptor-specific binding.