This AI system can analyse odor better than humans beings

It outperformed humans for 53 percent of the 400 compounds examined.
Jijo Malayil
Robot smelling a red rose
Robot smelling a red rose


When it comes to neuroscience, a crucial aspect of it is understanding how our senses translate light into sight, music into hearing, food into taste, and texture into touch. However, information regarding sensory relationships concerning smell has baffled researchers for a long. 

Humans regard the smell of flowers to be pleasant and the smell of decaying food to be offensive because of proteins in the nose called odor receptors. However, little is understood about how these receptors pick up chemicals and convert them into fragrances.

To understand the phenomenon, researchers from the Monell Chemical Senses Center and the Cambridge, Massachusetts-based startup Osmo, developed out of Google Research's machine learning initiative, examined the relationship between the brain's olfactory perception system and airborne chemicals. The study resulted in scientists having devised a machine-learning model that can now verbally describe the scent of compounds with human-level skill.

The details regarding their study are published in the journal Science. 

Extensive effort

There are around 400 active olfactory receptors in humans. These olfactory nerve proteins link with chemicals in the air to send a signal electrically to the olfactory bulb. According to the team, the number of olfactory receptors is far more than the four used for color vision or the 40 used for taste. 

“In olfaction research, however, the question of what physical properties make an airborne molecule smell the way it does to the brain has remained an enigma," said Joel Mainland, senior co-author and member of the Monell Center, in a statement. The team worked towards understanding the relationship between how molecules are shaped and how we ultimately perceive their odors. 

To that extent, the team created a model that could learn to correlate prose descriptions of a molecule's odor with the odor’s molecular structure. The map that results from these interactions generally consists of clusters of scents with comparable aromas, such as floral sweet and candy sweet.

A commercial dataset containing the molecular makeup and olfactory characteristics of 5,000 recognized odorants was used to train the machine. A molecule's form serves as the input for the algorithm, which predicts which odor words would best capture the molecule's aroma.

Furthermore, to ascertain the model's efficacy, researchers at Monell conducted a blind validation procedure in which a panel of trained research participants described new molecules and then compared their answers with the model’s description. The 15 panelists were each given 400 odorants and trained to use a set of 55 words - from mint to musty - to describe each molecule.

Impressive results

The AI model outperformed each panelist for 53 percent of the compounds examined in the research. The model also succeeded at olfactory tasks it was not trained to do. “The eye-opener was that we never trained it to learn odor strength, but it could nonetheless make accurate predictions," said Mainland.

The program was able to quantify a wide range of odor attributes, including odor intensity, for 500,000 possible scent molecules and find hundreds of pairings of structurally different compounds that had counterintuitively similar odors. “We hope this map will be useful to researchers in chemistry, olfactory neuroscience, and psychophysics as a new tool for investigating the nature of olfactory sensation,” said Mainland. 

The team hypothesizes that the model map may be set up according to metabolism, representing a significant change in how scientists see scents. In other words, smells perceptually similar to one another or nearby on a map are also more likely to have the same metabolic pathway. Currently, sensory scientists classify compounds like chemists would, for instance, by asking if a molecule has an ester or an aromatic ring.

According to researchers, the study will help the "world closer to digitizing odors to be recorded and reproduced. It also may identify new odors for the fragrance and flavor industry that could not only decrease dependence on naturally sourced endangered plants but also identify new functional scents for such uses as mosquito repellent or malodor masking."

What next for the team then? Find out how odorants mix and compete with one another to produce an aroma that the human brain perceives as having a completely distinct scent from each odorant.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board