Smell compounds classification using UMAP to increase knowledge of odors and molecular structures linkages

This study aims to highlight the relationships between the structure of smell compounds and their odors. For this purpose, heterogeneous data sources were screened, and 6038 odorant compounds and their known associated odors (162 odor notes) were compiled, each individual molecule being represented with a set of 1024 structural fingerprint. Several dimensional reduction techniques (PCA, MDS, t-SNE and UMAP) with two clustering methods (k-means and agglomerative hierarchical clustering AHC) were assessed based on the calculated fingerprints. The combination of UMAP with k-means and AHC methods allowed to obtain a good representativeness of odors by clusters, as well as the best visualization of the proximity of odorants on the basis of their molecular structures. The presence or absence of molecular substructures has been calculated on odorant in order to link chemical groups to odors. The results of this analysis bring out some associations for both the odor notes and the chemical structures of the molecules such as “woody” and “spicy” notes with allylic and bicyclic structures, “balsamic” notes with unsaturated rings, both “sulfurous” and “citrus” with aldehydes, alcohols, carboxylic acids, amines and sulfur compounds, and “oily”, “fatty” and “fruity” characterized by esters and with long carbon chains. Overall, the use of UMAP associated to clustering is a promising method to suggest hypotheses on the odorant structure-odor relationships.