QSAR Modeling of Sandalwood Odor

Researchers at the University of Vienna in collaboration with the University of North Carolina, have used Accelrys' QSAR program Tsar for Windows v3.3, to study the odor properties of a range of chiral α-campholenic derivatives.

The QSAR analysis accurately predicted the odor properties of a test set of molecules and highlighted the structural features necessary to create the sandalwood odor.

Chirality occurs frequently in nature and different chiral enantiomers of a molecule can exhibit very different properties. For example, Madrol, an α-campholenic derivative with a sandalwood odor exhibits either a sandalwood odor or a milky odor depending on the enantiomeric form isolated.

In this study, QSAR techniques have been used to correlate the molecular structure of various enantiomers with their sandalwood odor. A series of 44 α-campholenic derivatives were chosen for study. Of the 44 derivatives, 38 were randomly chosen for the training set and the remaining 6 were selected as the test set. The derivatives shared 12 common cores with 5 substituents defined for each core.

A major challenge was the selection of the odor property for prediction as the information came from different sources with each source using their own method of measurement. However, two scales were developed, see Table 1, one with equal intervals and the second with unequal intervals to increase differentation.

Odor intensity	Intensity levels
	Scale 1	Scale 2
Odorless	1	0.00
Very weak	2	0.05
Weak	3	0.20
Average	4	0.50
Strong	5	0.80
Very strong	6	0.95

Table 1 The scales used to rank the odor intensities. Scale 2 was introduced to increase the differentiation between weak, strong, and average.

The QSAR models were developed using Multiple Linear Regression (MLR) with leave-one-out cross-validation. Molecular descriptors were calculated with Tsar for Windows v3.3, and relevant ones were chosen by correlation analysis. On performing the analysis, it was discovered that 3 compounds were outliers and these were eliminated from the training set.

The same six descriptors appeared in the best models for both scales, see Table 2. However, the model for scale 2 - the unequal scale - was shown to be a better predictor of odor than the model built using scale 1. This is indicated in the statistical analysis of the two models. Model 1 had an R_o² of 0.86 and an R_o^'2 of 0.86 whereas model 2 had an R_o² of 0.95 and an R_o^'2 of 0.94.

Descriptor	Coefficient Model 1	Coefficient Model 2
Molecular volume (subst3)	-0.68	-0.13
Inertia moment 1 size (subst 4)	-720.54	-167.85
Dipole moment Z-component (subst 2)	-5.75	-1.06
Lipole X-component (subst 5)	1.61	1.07
Bond lipole (subst 2)	0.70	0.27
Group count for methyl (whole molecule)	243.63	0.13

Table 2 The descriptors used in each model and their coefficients.

The study also showed the importance of the substituents. For example, the interaction between a lipophilic substituent and the functionalised methyl group gives rise to the expression of the sandalwood odor. Also, descriptors relating to the methyl substituents on the 5 membered ring appear in both model 1 and model 2. They indicate that the number and position of the methyl groups is important.

This study has shown that QSAR can be used to generate an accurate model for odor prediction. Also, information about the role of substituents in the expression of sandalwood odor was obtained helping in the optimization of future sandalwood derivatives.

Reference

1. QSAR Modeling of α-Campholenic Derivatives with Sandalwood Odor, A. Kovatcheva, G. Buchbauer, A. Golbraikh, and P. Wolschann, J. Chem. Inf. Comput. Sci., 2003, 43, 259-266.