Researchers at Yale have developed a machine learning method that can predict how people perceive combinations of scents, an advance that could lay the groundwork for digitally mapping the human sense of smell. The study, published in the journal Proceedings of the National Academy of Sciences, addresses a long-standing gap: unlike vision and hearing, no quantitative framework exists for measuring how complex odors relate to one another.
Visual artists use color wheels. Audio engineers manipulate soundwave frequencies. But the smells people encounter daily - coffee, baked goods, body scent - are complex mixtures of dozens or hundreds of molecules, and researchers have lacked a metric for comparing them.
"Much progress was made over the past decade in identifying how individual molecules affect perceptions of odor, but the smells we encounter in the real world are composed of complex mixtures of dozens or hundreds of molecules, and we lack a metric for comparing and measuring them," said lead author Vahid Satarifard, a research scientist at Yale's Human Nature Lab. "This study shows that perceptual distances between odor mixtures can be accurately mapped, laying the groundwork for digitizing sense of smell."
How the model was built
The research grew out of a DREAM olfaction prediction challenge, an international competition organized by Pablo Meyer from IBM research. Teams were invited to develop predictive models that could measure perceptual similarities in scents produced by pairs of molecule mixtures.
Organizers compiled data from three prior studies on odor-similarity measurements into a unified dataset containing 168 unique single molecules, 731 unique molecule mixtures, and 507 mixture-pair measurements. Twenty-six teams competed over roughly three months to minimize prediction errors on a hidden test set of 46 scent mixture pairs.
After the competition, the Yale-led team built an ensemble model by averaging the predictions of the four top-performing models along with two other high-performing entries. Satarifard described the approach as "a machine-learning instance of the 'wisdom of crowds' phenomenon," where collective intelligence outperforms any single model.
The ensemble model beat existing state-of-the-art approaches and the top competition models on the hidden test set. It also maintained strong performance against an independent dataset of 50 newly designed scent-mixture pairs.
Language outperforms chemistry
One finding surprised the researchers: the highest-performing models relied heavily on semantic descriptions of odors rather than molecular structures.
"We found that using language features was very powerful in predicting similarity between two scent mixtures," Satarifard said. "This is interesting because English has a small vocabulary for smell, and we usually describe odors by naming objects. Something 'smells like flowers' or like watermelon or like a rotten egg, whereas colors have specific names like 'green' or 'blue.' That was thought to make semantic odor descriptors a poor basis for predicting how smells relate, but we found the opposite."
The result contributes to an ongoing discussion about whether language, rather than chemical composition, offers a better route to describing and modeling smell.
Potential health applications
The researchers see practical applications on the horizon, particularly in health monitoring. Many conditions, including Parkinson's disease and cancer, have been reported to carry odor signatures.
"One ultimate goal here is to reach a place where we can use odor as a disease biomarker," Satarifard said. "There could be technology several years from now that can monitor your smell and alert you to any changes that might require additional screening."
Nicholas Christakis, Sterling Professor of Sociology and Natural Science at Yale and director of the Human Nature Lab, pointed to a non-clinical use as well. "Body scent is also a complex mixture of odors," he said, "and we suspect that it plays an important role in human social interactions."
Why this matters for science and research professionals
For researchers working in sensory science, computational biology, or health diagnostics, this study provides a validated framework for quantifying odor perception - a domain that has resisted the kind of systematic measurement common in vision and acoustics. The ensemble modeling approach also demonstrates a reproducible method for combining competing predictive models, a technique with applications beyond olfaction. As odor-based diagnostics move toward clinical use, researchers in biomedical fields should watch for datasets and benchmarks emerging from this line of work.
Your membership also unlocks:
