Computer-aided aroma design. II. Quantitative structure–odour relationship

Computer-aided aroma design (CAAD) is likely to become a hot issue as the REACH EC document targets many aroma compounds to require substitution. The two crucial steps in computer-aided molecular design (CAMD) are the generation of candidate molecules and the estimation of properties, which can be difficult when complex molecular structures like odours are sought and their odour quality are definitely subjective or their odour intensity are partly subjective as stated in Rossitier's review [K.J. Rossitier, Structure–odour relationships, Chem. Rev. 96 (1996) 3201–3240]. The CAAD methodology and a novel molecular framework were presented in part I. Part II focuses on a classification methodology to characterize the odour quality of molecules based on structure–odour relation (SOR). Using 2D and 3D molecular descriptors, linear discriminant analysis (LDA) and artificial neural network are compared in favour of LDA. The classification into balsamic/non-balsamic quality was satisfactorily solved. The classification among five sub-notes of the balsamic quality was less successful, partly due to the selection of the Aldrich's catalog as the reference classification. For the second case, it is shown that the sweet sub-note considered in Aldrich's catalog is not a relevant sub-note, confirming the alternative and popular classification of Jaubert et al. [J.N. Jaubert, C. Tapiero, J.C. Dore, The field of odour: toward a universal language for odour relationships, Perfumer Odourist 20 (2) (1995) 1–16], the field of odours.