A QSAR Study on a Set of 105 Flavonoid Derivatives Using Descriptors Derived From 3D Structures
BioChem Press, Marjan Vra ko, Johann Gasteiger · 2002
A relationship between the 3D structure and biological activity was studied for a set of 105 flavonoid derivatives using a counterpropagation neural network. The 3D structures were determined in two ways, either by the empirical structure generator CORINA or by optimization within the semiempirical AM1 approximation. Furthermore, we compared two types of structure representations, the radial distribution function (RDF) method and the ‘spectrum–like’ representation. We show how different methods for 3D structure determination and different representations influence the quality of QSAR models. For all methods considered we found comparable models for the relationship between structure and biological activity. The computation times in 3D structure determination are visibly shorter for CORINA as against the AM1 method.