Scientist have moved closer to digitizing the sense of smell. This is based on a new computer model that is capable of interpreting and describing odours better than human panellists.
Smell is a complex sensory relationship. To better understand the mechanisms at play, question, scientists at the Monell Chemical Senses Center are investigating how airborne chemicals connect to odour perception in the brain.
To smell, humans use about 400 functional olfactory receptors. These are proteins located at the end of olfactory nerves that connect with airborne molecules to transmit an electrical signal to the olfactory bulb. However, exactly what physical properties make an airborne molecule smell the way it does to the brain has remained an enigma.
This led to the question: Can a computer discern the relationship between how molecules are shaped and how we ultimately perceive their odours? If so, this could provide the basis for understanding of how our brains and noses work together. To answer the question, the researchers developed a machine learning model that learned how to match the prose descriptions of a molecule’s odour with the odour’s molecular structure.
The development phase involved training the artificial intelligence by using an industry dataset that included the molecular structures and odour qualities of 5,000 known odorants. The data input was the shape of a molecule, and the output is a prediction of which odour words best describe its smell.
To test out the model’s effectiveness, a panel of 15 participants were each given 400 odorants. The group were trained to use a set of 55 words — from mint to musty — to describe each molecule.
For example, taking one of the odours – a previously uncharacterized odorant 2,3-dihydrobenzofuran-5-carboxaldehyde – the consensus was that the odour was very powdery (5 out 5 on the scale) and somewhat sweet (3 out of 5 on the scale).
Through this, the researchers have discovered that a machine-learning model has achieved human-level proficiency at describing, in words, what chemicals smell like. In comparing the model’s performance to that of individual panellists, the model achieved better predictions of the average of the group’s odour ratings.
Specifically, the model performed better than the average panellist for 53 percent of the molecules tested.
It is possible the model may identify new odours for the fragrance and flavour industry. A consequence of this could be to decrease humanity’s dependence on naturally sourced endangered plants.
Other applications could include the identification of new functional scents for such uses as mosquito repellent or malodour (unpleasant smell) masking.
The research appears in the journal Science, titled “A principal odor map unifies diverse tasks in olfactory perception.”
