Researchers from the University of CA, Riverside, trained an artificial intelligence (AI) algorithm to detect a chemical's smell and predict if a human will like that smell. 80 percent of flavor comes from odors and if the fragrance isn't attractive, consumers will avoid buying that product. Using AI, food and fragrance companies can find attractive replacements for unpleasant-smelling chemicals.

Anandasankar Ray is a professor of molecular, cell and systems biology at UC Riverside.Source: L. Duka.

Humans sense odors through over 400 odorant receptors (OR) in the nose. ORs are activated by chemicals and can detect a vast number of chemicals. The team modeled human olfactory percepts using chemical informatics and AI.

The AI algorithm evaluated a large number of chemical features to learn what a chemical smells like and predict its attractiveness. This could be used to prioritize the chemicals used in food, flavor and fragrance. With the algorithm, these industries could find replacement chemicals for rare or expensive chemicals and chemicals that humans think smells bad.

The team screened about half a million compounds for ligands that activate 34 odorant receptors. They then focused on if the algorithm can estimate OR activity and predict the qualities of odors. They used 100s of chemicals that were previously evaluated by humans to teach the AI system what humans like and don’t like to smell.

The algorithm-based ORs accurately predicted 146 percepts of chemicals. Only a few ORs were needed to predict some of the smells. When tested further in fruit flies, the team saw the same promising results when that algorithm was tasked with predicting fly attraction and aversion to odorants.

A paper on this research was published in iScience.