Analysis of Odor Structure Relationships Using Electronic Van Der Waals Surface Property Descriptors and Genetic Algorithms

Abstract

This chapter describes a new odor structure relationship (OSR) correlation methodology that utilizes large olfactory databases available in the open scientific literature as input. The first step in this procedure is to represent each molecule in the database by an appropriate set of molecular descriptors. To accomplish this task, Breneman's Transferable Atom Equivalent (TAE) descriptor methodology is used to create a large set of electron density derived shape/property hybrid descriptors. These descriptors have been chosen because they con-elate with key modes of intermolecular interactions and contain pertinent information about shape and electronic properties of molecules. In contrast to more traditional methodologies that have shown not to be effective, our use of shape-aware electron density based molecular property descriptors has eliminated many of the problems associated with the use of descriptors based on substructural fragments or chemical topology. A second reason for the limited success of past OSR efforts can be traced to the complex nature of the underlying modeling problem. To meet this challenge, we have developed a genetic algorithm for pattern recognition analysis that selects descriptors, which create class separation in a plot of the two or three largest principal components of the data. Because principal components maximize variance, the bulk of the information encoded by these descriptors is about differences between the odorant classes in a data set.