Analysis of a system of description of odors by means of four different multivariate statistical methods

Abstract

Abstract. In order to analyze the relationships among 32 descriptors of odors (notes), similarity coefficients were calculated using a data bank of 628 odoriferous products. The simple examination of the similarity matrix (32,32) has shown notes selectively and strongly associated (e.g. camphoraceous -pincy and musky-powdery) and others less selectively associated (e.g. floral, green and herbaceous). This analysis was completed by four multivariate statistical methods. Non-linear mapping (NLM) proved to be more efficient than principal coordinates analysis for planar representation of olfactory notes, and has given results similar to those previously obtained using other data and other methods (similar disposition of notes around the central note 'floral'). Furthermore, the ascending hierarchical taxonomy and the minimal spanning tree were coherent with the NLM representation. These three methods complete each other and constitute a convenient system to analyze odor descriptions.

Underlying dimensions in the descriptive space of perfumery odors: Part II

“…These similarities, which are well known in perfumery (Chastrette et al, 1988(Chastrette et al, , 1991, are reflected in the loading plots shown in Fig. 1 because pairs of attributes strongly associated appear close to each other.…”

“…In most cases, the interpretation of results becomes more straightforward when the matrix columns are mean-centered prior to applying PCA. In this case, given that S Abe and S SAFC are symmetric arrays, they were double-centered using the procedure applied by Chastrette et al (1991) (Eq. 2), which leads to transformed matrices suitable for PCA (Gower, 1966) that will be referred to as T Abe and T SAFC , respectively:…”

“…Such approach has been carried out by different researchers (Abe et al, 1990;Chastrette et al, 1991;Jaubert et al, 1995;Madany-Mamlouk et al, 2003;Zarzo & Stanton, 2006). However, the sensory maps obtained in these works are difficult to compare.…”

Understanding the underlying dimensions in perfumers' odor perception space as a basis for developing meaningful odor maps

“…Different researchers have numerically coded the semantic descriptions contained in these databases and have analyzed them with multivariate statistical methods (Abe, Kanaya, Komukai, Takahashi, & Sasaki, 1990;Chastrette, de Saint Laumer, & Sauvegrain, 1991;Chastrette, Elmouaffek, & Sauvegrain, 1988;Jaubert, Tapiero, & Doré, 1995;Madany-Mamlouk, Chee-Ruiter, Hofmann, & Bower, 2003;Zarzo & Stanton, 2006). The results led to the classification of descriptors into 10 to 32 clusters, confirming the belief that olfactory perception space is high-dimensional.…”

“…Similarly, the Flavornet database (www.flavornet.org) classifies odor descriptors into 25 categories. Another outcome of some of these studies (Abe et al, 1990;Chastrette et al, 1991;Jaubert et al, 1995;Madany-Mamlouk et al, 2003) is the development of two-dimensional representations of odor descriptor space. However, none of them provides a meaningful interpretation for the two dimensions that make up the odor map.…”