Algorithm screens through hundreds of millions of molecules to make it easier to replicate any given smell
Hundreds of billions of molecules with odors exist. If you want to create a particular scent or flavor, say char-grilled beef, rose petals or freshly baked bread, you’d need to screen thousands or millions of compounds to match what you were looking for. It would take thousands of hours and hundreds of thousands of dollars in human testing.
An Arizona State University scientist helped create an algorithm that winnows down 99 percent of that workload.
The algorithm doesn’t find a needle in a haystack. It finds the handful of hay with the needle in it. It's one of the latest developments from the school selected as the nation's most innovative university by U.S. News & World Report for two years straight. The potential applications are boundless, highlighting the freedom researchers at ASU have to track down big-picture projects.
Rick Gerkin, an assistant research professor in the School of Life SciencesThe School of Life Sciences is part of the College of Liberal Arts and Sciences., investigates how smell perception, learning and behavior are represented in the brain. He explained what the algorithm he co-wrote does.
“If you wanted to find a molecule that fit that smell, by taking samples off the shelf and testing them, that would take a long time,” said Gerkin (pictured above). “What we have is a computational approach that can screen through any molecule in existence and can tell you it’s a candidate for that smell. … If you had 100,000 molecules, it’d be great if we could get you the one, but we narrow it down to 1,000.”
Why things smell the way they do is poorly understood, according to Gerkin. It’s a wicked problem for the flavor and fragrance industries. International Flavors and Fragrances, an industry leader, spends $250 million a year on research and development, Gerkin said.
Gerkin and his co-authors had collected a lot of data based on extensive smell-testing of 49 human subjects asked to sniff 476 different odor chemicals. Subjects were asked to tell how pleasant the odor was, how strong the odor was, and how well the smell matched a list of 19 descriptors. Gerkin’s team decided they wanted to reach out to the community and see who could predict human olfaction.
They submitted the data set to the DREAM Challenges, a collaborative open science effort made up of researchers from academia, technology, industry and nonprofits, which focuses on biological and biomedical research problems. Challengers will put out a data set on a subject like mammography or genomics and then see who can make sense of it. They give out some of the data and ask scientists to build a model that can predict the rest.
“If you predict the data, you’ve done a great job,” Gerkin explained. Of the 28 labs that participated in the challenge, Gerkin’s had the best model.
“We proved that the model is nearly as good as can possibly be made, given the data available,” he said. “In other words, an oracle with access to the same data couldn't really make a much better model.”
The model makes it possible to prescreen all possible molecules and select a tiny fraction — as few as a few hundred — for testing.
There are probably hundreds of billions of molecules that exist that have an odor.
“That’s the space we’re talking about,” Gerkin said. “This can cut the workload by 99 percent and make it practical to possibly discover molecules that evoke any desired percept, and then to put those to use commercially.”
Whether and how soon industry adopts the math is up in the air. He has talked to International Flavors and Fragrances. “The thing is, it’s an algorithm,” he said. “They’re interested in it, but it’s an open-source kind of thing. They can implement it however way they’d like to. … What they would probably want to do is use the basic model, but collect their own data.”
Gerkin said the algorithm could bring ideas like Smell-O-Vision to (virtual) reality. Smell-O-Vision was a system that released odors in movie theaters so viewers could smell what was going on in the film. It was used for only one movie in 1960 before being abandoned.
However, hunting virtual dinosaurs and smelling the jungle around you in a game would be compelling, Gerkin said.
“I think it would be very emotional,” he said, “to have smell as part of your experience.”
Rick Gerkin, an ASU assistant research professor, has helped create an algorithm that will make a big stink in the world of smells. Photo by Charlie Leight/ASU Now