Aim: The quantitative structure-activity relationship approach was applied to understand the relative binding affinity of triphenyl acrylonitriles to estrogen receptors.
Material and methods: A sample of previously studied triphenyl acrylonitriles was divided into training (18 compounds) and test sets (7 compounds) using a stratified random approach. The molecular descriptor family on vertices cutting (MDFV) approach was used in order to translate the structural information into descriptors. The relationship between binding activity and structural descriptors was identified using the multiple linear regression procedure.
Results: An optimal three-parameter equation with a determination coefficient of 0.9580 and a cross-validation leave-one-out parameter of 0.9408 was identified. The optimal model was assessed on a test set and a determination coefficient of 0.9004 was obtained. The MDFV model proved not to be significantly different from the previously reported model in terms of goodness-of-fit. In terms of information criteria (Akaike's, Bayesian, Amemiya, and Hannan-Quinn) and Kubinyi function, the MDFV model proved to perform better than the previously reported model.
Conclusion: The optimal MDFV model was able to explain approximately 96% of the total variance in the estrogenic binding relative affinity of triphenyl acrylonitriles and to have estimation and prediction abilities. Although there were no significant differences in terms of goodness-of-fit, the MDFV model proved to exhibit better information parameters compared to the previously reported model using the same number of molecular descriptors.