Statistical Modeling of Molecular Shape, Similarity, and Mechanism

Abstract

Current methods of molecular shape description and comparison have two major shortcomings: they are relatively nondirectional, and the superposition of molecules is usually a subjective procedure. A new method of shape description and comparison is described, which uses a binding moment, derived from fragment binding constants, to orient the molecules, followed by SIMCA modeling of the shape and similarity of the structures. This method is compared with other molecular shape descriptors of varying dimensionality. The shape descriptors are correlated with acute and chronic lung and liver toxicities of a set of pyrrolizidine alkaloids. The overall results, although not precise enough for quantitative prediction, demonstrate the value of using more specific and directional shape descriptors, coupled with a more rational method for molecule orientation and superposition.