ERROR 1
ERROR 1
ERROR 2
ERROR 2
ERROR 2
ERROR 2
ERROR 2
Password and Confirm password must match.
If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)
ERROR 2
ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.
A new map charts a landscape of smell and can predict odor based on a molecule’s structure (Science 2023, DOI: 10.1126/science.ade4401).
Aromas are tricky to quantify and categorize. In vision, colors can be identified by their characteristic light wavelengths. “In olfaction, you have no single axis like that to organize all the molecules on,” says Joel D. Mainland, a neuroscientist at Monell Chemical Senses Center.
Structural similarity between molecules doesn’t necessarily predict smell similarity. And attempts to build models linking functional groups or molecular footprints to perceived odors have come up short in mapping the odor space. So Mainland’s team turnedto graph neural networks, a type of machine learning that’s well suited to molecules. Graph neural networks use a data structure that captures intel from molecules’ structures, such as each atom’s valence, charge, hydrogen count, and hybridization, and each bond’s degree and aromaticity.
From two databases, the researchers curated descriptions—labels such as fruity, grassy, or cheesy—for around 5,000 molecules. They fed these descriptions, along with information drawn from the molecules’ structures, to the neural network for training. Then the neural network selected, which attributes matter for predicting a molecule’s smell, creates a colorful 2D map that organizes the odors.
The team then tested how the model faired at predicting the smells of some 400 molecules it hadn’t yet seen. The researchers compared the model’s output with that of a panel of trained human sniffers. For most of the molecules, the model’s assessment of the smell was more reliable than that of a typical panelist.
Once you can quantify odors, you can use that information for communication, Mainland says. For example, some people can detect a smell associated with Parkinson’s disease. Mapping that smell could be useful for diagnosis. Molecules could be predicted for desired applications, such as making crops smell yummier, bringing a certain note to a fragrance, or masking the stench of used kitty litter.
“While this study presents an impressive milestone, the model is far from perfect,” says Michael Schmuker, a neuroscientist and olfaction expert at the University of Hertfordshire. It’s limited by individual variability in scent perception, he says. Schmuker hopes this work prompts further research into what causes this variability, such as differences in environment and genetics.
The study also opens up the potential to gain new insights on why people are tuned into certain smells, says Tatyana O. Sharpee, a computational neuroscientist at the Salk Institute for the Biological Sciences who wasn’t involved with the work. Researchers could use the model to investigate the biological reasons people might tune into odorants present at low concentration.
Aromas in the real world generally stem from multiple molecules. Mainland and colleagues hope to extend their work into mixtures, perhaps finding a set of primary odors—similar to the idea of primary colors—that make up much of what we smell, he says.
Join the conversation
Contact the reporter
Submit a Letter to the Editor for publication
Engage with us on Twitter