Novel Classification of Mono-Molecular Odorants using Standardized Semantic Profiles

Odorants are typically classified by specially trained individuals using subjective verbal scent descriptors. Herein, we used natural language processing to develop standardized semantic profiles of mono-molecular odorants. We have (i) curated and integrated scent perception data for mono-molecular odorants from 4 online sources; (ii) represented verbal scent descriptors used in these sources as vectors in semantic space; (iii) calculated average semantic distances between vectors representing each mono-molecular odorant and each of the vectors for a set of 27 standard verbal scent descriptors to yield 27-dimensional harmonized odorant semantic profile; and (iv) applied dimensionality reduction techniques to these harmonized profiles, to visualize clustering of odorants with similar semantic profiles. This novel uniform representation of odorants can be employed to transform any subjective verbal description of any odorants into standardized semantic profiles that can facilitate automated classification, structure-odor relationship studies, and design of odorants with the desired scent.