QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIP STUDIES USING ARTIFICIALNEURAL NETWORKS

The potential of using back propagation (BP) type artificial neural ne twork for correlation of biological activity with structural and physi cochemical descriptors of chemical compounds has been demonstrated. Th e datasets studied concern aromatic and heteroaromatic nitro compounds , arylhydroxamic acids and different hexestrol derivatives. The biolog ical activities like mutagenicity, in vitro 5-lipoxygenase inhibitory potency and estrogen binding affinities have been correlated with the physicochemical parameters such as energy of LUMO, hydrophobicity and Hammett's constant and/or van der Waals' volume. Better quantitative f it could be obtained compared to previous works using simpler networks and less computer time. The capabilities of statistical and artificia l neural network methods for QSAR studies have been compared.