Scientists Stink at Reverse-Engineering Smells

But artificial intelligence could finally help researchers tell a molecule’s odor from its structure.

"Smells of tomorrow's regret." (Stefano Rellandini / Reuters)

Light and sound are predictable. Smells are not.

If you knew the wavelength of a beam of light, you could tell me what most people would see when they looked at it: 480 nanometers looks blue, and 650 nanometers looks red. If you knew the frequency of a musical note, you could name that note: 261 Hertz is middle C.

But if you saw the chemical structure of a molecule, you wouldn’t know what it smelled like—or even if it smelled of anything at all. Unless you actually stick your nose over some benzaldehyde, you wouldn’t be able to predict that it smelled like almonds. If you saw dimethyl sulfide drawn on a page, you couldn’t foresee that it carried the scent of the sea.

This is a longstanding problem, but one that a team of scientists—and a horde of volunteers and citizen scientists—have come a little closer to cracking. Through a crowdsourced competition, Andreas Keller and Leslie Vosshall at Rockefeller University and Pablo Meyer at IBM have developed algorithms that can reverse-engineer the smell of a molecule—to predict what it smells like from what it is.

These virtual noses are far from perfect, but they’re much better than anyone had thought possible. And they’re a first step towards reverse-engineering smells, by designing molecules that smell a certain way—a feat that would be a huge boon for the perfume and flavor industries.

“I think a lot of people would have said that this problem is impossible to crack,” says Vosshall. “The fact that we’ve made any progress at all was surprising.”

This has been hard, she says, for two reasons. First, it's hard to know which aspects of a molecule contribute to its odor. We know that the wavelength of light determines its color—simple. But a molecule’s smell might depend on the number of carbon atoms it has, how stable it is, and the chemical branches that protrude from it. Second, scientists who study smell have understandably focused on molecules that are relevant to the food and perfume industries. That’s like trying to understand color vision by only studying red, and ignoring blue and green. “It’s a cramped space,” she says. “We tried to break out.”

She and colleague Andreas Keller began by collecting a much broader range of 480 molecules—including unfamiliar, unpleasant, and even odorless ones. They then presented these chemicals to 55 volunteers, whom they recruited through Craigslist. The participants visited the lab and worked their way through rack upon rack of glass vials, opening and inhaling. They noted whether they smelled anything at all. If so, they rated the scents along several categories. How intense or pleasant is it? How garlicky? Fishy? Fruity?

The result was one of the largest data sets on smell ever collected, which Keller and Vosshall turned over to Meyer at IBM. He runs the DREAM Challenges, in which volunteers compete to build machine learning algorithms that can make useful predictions from large sets of data. “I like to call it the piñata of science—everyone is trying to hit the problem with a different algorithm,” Meyer says. The competitors have tackled everything from improving the accuracy of mammograms to predicting how people respond to cold viruses. And between January and May 2015, 22 teams of them worked on smells.

Richard Gerkin from Arizona State University and Yuanfang Guan from the University of Michigan led the teams that produced the winning algorithms, which were constructed using data from 338 molecules, refined using 69 more, and finally tested on another 69. After the challenge, the teams shared all their results to produce even better models.

On a performance scale from 0 to 1, their ability to predict a molecule’s smell came in at 0.71 for pleasantness, 0.78 for intensity, and anywhere from 0.1 to 0.7 for the other 19 descriptors like garlic, fish, fruit, sour, musky, decayed, sweaty, sweet, grassy, grass, and burnt. As the team writes, “It [is] possible to predict the perceptual qualities of virtually any molecule with an impressive degree of accuracy to reverse-engineer the smell of a molecule.”

Their scores may not sound like much, but they’re better than those from previous studies. “They’re not blockbuster in terms of how correlations go, but having done some work like this myself, I know that’s actually quite impressive,” says Jason Castro from Bates College.

Human noses aren’t perfect either, he adds. “If I give you a chemical to smell on Tuesday, and you smell it again on Wednesday, there’s going to be variability in your perception. These models are doing as well as that intrinsic human variation.” There’s also variation between people. You and I might not smell the same molecule in the same way, which is why the DREAM algorithms were reasonably at predicting what molecules smell like on average, but less good at predicting what they smell like to specific volunteers.

It’s progress, but “we haven’t solved the problem,” says Vosshall. There is still no neat theory that pairs a molecule’s characteristics with its odor. A molecule’s smell might depend on the number of carbon atoms it has, how stable it is, and the chemical branches that protrude from it. The algorithms hint at certain features that correlate with certain qualities: The presence of sulphur atoms makes things more likely to smell burnt or garlicky. Bigger molecules are more likely to smell pleasant. Molecules that are similar to vanillin smell more like baked goods.

“We haven’t been able to find a tight correspondence between Feature X and Perceptual Quality Y,” says Castro. “There are so many potential ways in which molecules can differ.”

Machine learning can help to cope with that complexity, by identifying important features faster than any human could. But that only works if you have a good set of data to begin with. And even Vosshall and Keller’s mammoth data set was just a start. They now want to test even more volunteers on an even wider range of smells. And they want to get people to make judgements about similarity—that is, if I give you one molecule, how similar do you think these other ten smell?

That will mean getting many recruits to spend days hunched over racks of vials—a somewhat thankless task, but one that people have an oddly large appetite for. For their first study, “each of these hardworking and dedicated members of the public put in 20 hours, and we ended up having to turn people away,” Vosshall says. “There’s something really seductive that draws people into olfactory studies. Smell makes us think about sex and perfume and food. Smell is sexy.”