Electronic Nose Knows a Good Smell

Researchers have trained an "electronic nose" to distinguish between sweet and foul odors, even if it's never been exposed to them before. The advance could help companies better monitor for dangerous chemicals and spoiled foods, and could even allow odors to be transmitted electronically. The work also challenges the idea that people from different cultures have different scent preferences.

Artificial noses have been around for more than 20 years. These devices typically consist of arrays of chemical sensors that respond to odorant molecules. Each sensor responds slightly differently to a given odor, and taken together, the sensors give an “odor fingerprint”—a characteristic response pattern, such as a series of colors to each odor.

Most of these devices have been able to identify and distinguish only between specific odors they've previously been trained to recognize, however, says neuroscientist Rafi Haddad of the Weizmann Institute of Science in Rehovot, Israel. If an artificial nose is ever to replace the real thing, he says, it will have to be able to classify odors it has never encountered before.

Toward that end, Haddad and his colleagues wanted to see if they could train an electronic nose, or eNose, to evaluate the quality of novel odors. They first presented 76 odors to an array of 16 eNose sensors and recorded changes in the conductivity of the polymers that made up the sensors. Then they fed these data into a computer along with information about how human volunteers had classified the odors—ranging from “the best odor you have ever smelled” to “the worst odor you have ever smelled.”

After this calibration phase, the researchers introduced 22 additional odors both to the eNose and to a new group of volunteers. The eNose and the humans agreed on the relative pleasantness of these odors more than 80% of the time, the team reports today in PLoS Computational Biology. In another trial, the researchers presented only the odors rated most pleasant or most unpleasant by human volunteers and asked the eNose to classify each. In this experiment, it agreed with humans more than 90% of the time.

The researchers also tested the eNose on two groups of people: native Israelis and native Ethiopians. Previous work has suggested that at least some olfactory perceptions vary between cultures. But Haddad and his colleagues found that an eNose calibrated to agree with Israeli scent preferences predicted Ethiopians' preferences equally well, suggesting that that our preference for particular scents may transcend culture.

Showing that an artificial nose can predict how humans will perceive smells is “a good result,” says computational neuroscientist Barani Raman of Washington University in St. Louis, although it remains unclear what the eNose is responding to at a molecular level. “It would be very interesting to see what molecular properties correlate with pleasantness and what these sensors actually detect,” he says.

There are other subtleties in odorant molecule detection that are difficult to explain, Raman adds. For example, the chemical skatole is found at a high concentration in feces, and both the eNose and the human subjects considered it a “very unpleasant” odor. But at low concentrations, skatole smells like flowers: It's found naturally in orange blossoms and jasmine and is used as flavoring in some ice creams. Explaining how such perceptual differences arise—from either biological or artificial noses—is “a very complex problem,” Raman says.