How the experimental design associated with objectivized sensory analysis can be used to predict odor quality of gaseous mixtures?

From environmental nuisances to perfumes and aromas, the need to understand odorous interactions is growing. In this context, sensory analysis is often used but faces an obstacle when dealing with odor mixtures. Indeed, sensory approaches provide data on the global nature and intensity of these mixtures, but these methods cannot reveal the various interactions taking place when perceiving an odor. To address this issue, modeling the vari-ations in odor nature and intensity according to the composition of complex mixtures offers insight into sensory interactions as well as a tool to predict odor nature and intensity. While several models have been developed for odor intensity, models for odor nature are scarce. In this study, an experimental mixture design is combined with sensory analysis with the aim of developing new numerical models that transform chemical data into sensory data. Mixtures of four odorants, dimethyl disulfide (DMDS), toluene, furfuryl mercaptan, and cyclopentanone are composed of multiple values of the Odor Activity Value (OAV) for each odorant. The odor nature of each mixture is described using Langage des Nez (R) (LdN), an objectivized odor nature description method. The variation in odor nature with the composition of the mixture is modeled using the experimental mixture design. The yielded models are shown to be valid and of relatively good descriptive (adjusted R2) between 0.610 and 0.925) and predictive (predictive R2 between 0.390 and 0.886) qualities. The models provide insight into the individual contributions and binary and ternary interactions among the odorants impacting the global odor nature. Moreover, the models allow the odor nature of a mixture to be predicted by transforming the chemical con-centration of each odorant into an odor nature description. These results may be an answer to the question raised by the odor mixture effect.