Odor-structure relationship studies of tetralin and indan musks.

A list of 147 tetralin- and indan-like compounds was compiled from the literature for investigating the relationship between molecular structure and musk odor. Each compound in the data set was represented by 374 CODESSA and 970 TAE descriptors. A genetic algorithm (GA) for pattern recognition analysis was used to identify a subset of molecular descriptors that could differentiate musks from nonmusks in a plot of the two largest principal components (PCs) of the data. A PC map of the 110 compounds in the training set using 45 molecular descriptors identified by the pattern recognition GA revealed an asymmetric data structure. Tetralin and indan musks were found to occupy a small, but well-defined region of the PC (descriptor) space, with the nonmusks randomly distributed in the PC plot. A three-layer feed-forward neural network trained by back propagation was used to develop a discriminant that correctly classified all the compounds in the training set as musk or nonmusk. The neural network was successfully validated using an external prediction of 37 compounds.