Chemometrical treatment of electronic structures of 28 flavonoid derivatives

Electronic structures of 28 flavonoid derivatives were treated with the che mometrical methods, i.e., multiple linear regression, counterpropagation ar tificial neural networks, principal component analysis, and partial least-s quares method. Electronic structures given as density-of-states spectra wer e constructed from calculated molecular orbital energies. It was shown that all the methods recognize some structural features from electronic structu res. The relationship between electronic structures and biological activity was also studied. The reported results, i.e., the predicted activities, ar e comparable to the results obtained by models built up with standard descr iptors, i.e., topological indices, geometric parameters, and electrostatic indices or "spectrum-like representations of three-dimensional structures." For Linear models the correlation coefficients obtained by a cross validat ion leave-one-out method are close to 0.9, and for artificial neural networ k models they are about 0.8. (C) 2000 John Wiley & Sons, Inc.