Scientists have made a set of algorithms to predict smells based on their molecular structure.
Of our five senses, it could be argued that we rely on sight more than the others. Yet, sight is fairly straight forward: each wavelength of light in the visible spectrum corresponds to a color. Smell is more complicated; involving chemical compounds and molecules which all have to be interpreted. As such, making a device which reads light or color is fairly straight forward as well, but making a machine that can smell is a different task entirely. A team of 22-computer scientists have just created such a device, which analyzes the structure of molecules to accurately determine what they smell like.
The smell prediction algorithm is the result of a previous study by Leslie Vosshall and colleagues at The Rockefeller University in New York City. Vosshall’s research team started by asking 49 volunteers to rate almost 500 different kinds of smells from different vials. The participants were asked to rate the smells by intensity, pleasantness, and by labeling them with one of 19 different descriptors such as “burnt”, “garlic”, and “fish”. The study created a database of over a million data points.
Computational biologist Pablo Meyer learned of the Rockefeller study two years ago and realized that the data points could serve as an opportunity. Meyer works for IBM’s Thomas J. Watson Research Center in Yorktown Heights, New York, but also heads the DREAM challenges, contests that ask teams of computer scientists to solve outstanding biomedical problems.
“I knew from graduate school that olfaction was still one of the big unknowns,” Meyer says. While researchers have discovered some 400 different smell receptors among humans, understanding the logic and function behind smell is something we haven’t quite worked out yet. After hearing of the Rockefeller study, Meyer set up a DREAM challenge to build a machine that could predict smells using the study’s data points. Twenty two teams took up the challenge, but two stood out – A team led by Yuanfang Guan, a computer scientist at the University of Michigan in Ann Arbor, was best at predicting how an individual would rate a smell. Another team led by Richard Gerkin at Arizona State University in Tempe was best at guessing how smells were rated on average.
“We learned that we can very specifically assign structural features to descriptions of the odor,” Meyer says. Molecules with sulfur groups tend to produce a “garlicky” smell, and molecules with a chemical structure similar to vanillin tends to produce a “bakery” smell. What practical use this technology will have remains to be seen.