Multivariate statistical analysis of a large odorants database aimed at revealing similarities and links between odorants and odors

The perception of odor is an important component of smell; the first step of odor detection, and the discrimination of structurally diverse odorants depends on their interactions with olfactory receptors (ORs). Indeed, the perception of an odor's quality results from a combinatorial coding, in which the deciphering remains a major challenge. Several studies have successfully established links between odors and odorants by categorizing and classifying data. Hence, the categorization of odors appears to be a promising way to manage odors. In the proposed study, we performed a computational analysis using odor descriptions of the odorants present in Flavor-Base 9th Edition (2013). We converted the Flavor-Base data into a binary matrix (1 when the odor note appears in the odor description, 0 otherwise). We retained 251 odor notes and 3508 odorants, considering only the orthonasal perception. Two categorization methods were performed: agglomerative hierarchical clustering (AHC), and self-organizing map (SOM). AHC was based on a measure of the distance between the elements performed by multidimensional scaling (MDS) for the odorants, and correspondence analysis (CA) for the odor notes. The results demonstrated that the SOM classes appeared to be less dependent on the frequency of the odor notes than those of the AHC clusters. SOMs are especially useful for identifying the associations between less than 4 or 5 odor notes within groups of odorants. The obtained results highlight subsets of odorants sharing similar groups of odor notes, suggesting an interesting and promising way of using computational approaches to help decipher olfactory coding.