Radial basis function networks in host–guest interactions: instant and accurate formation constant calculations

Abstract

The application of the second most popular artificial neural networks (ANN), namely, the radial basis function (RBF) networks, have been developed for obtaining sufficient quantitative structure-formation relationships (QSFR) with improved accuracy. To this end, a data set of 17 barbiturates as guests complexing to α- and β-cyclodextrins (CDs) was examined using RBF and generalized regression neural networks (GRNN) as function approximation systems. The proposed methods led to substantial gain in both the prediction ability and the computation speed of the resulting models compared to regression models. For the development and evaluation of the ANN systems, the same (four) descriptors used by Lopata in a former study [A. Lopata, J. Pharm. Sci. 25 (1985) 777–784] were used also in the present study. Some of the proposed models diminished substantially the number of outliers, during their implementation to unseen (new) barbiturates.