Although the Roman philosopher Lucretius was right when he wrote that odors were caused by a flow of atoms emitted by objects, smell may still be the least understood of our five senses. While we use it every day, science has not fully understood how molecules produce an odor, or how to determine what they smell like without sniffing them. We have known for decades that for sight, seeing requires wavelengths of light, and for sound, hearing is done through tonal frequencies, but the olfactory-code remained unbroken – until now.
As the organizers of the DREAM Olfaction Prediction Challenge, we have published results in Science demonstrating, for the first time, that it is possible to predict odor from molecular structures. In this publication, our team explains our capability to build a model to predict the odor qualities of a certain molecule, just by using its structure. Such a model will provide fundamental insights into how chemicals are transformed into a smell percept in the brain; how the brain perceives odors. Beyond this fundamental biological understanding, the ability to reverse-engineer smells by designing molecules would be a major technological advance for perfume and flavor industries.
More than twenty global teams from academia and other corporate research organizations accepted the challenge and worked to understand how to predict smells. To start, challenge participants used the world’s largest olfactory psychophysical dataset, created by our colleague Andreas Keller at Rockefeller University. This data was drawn from 49 individuals who sniffed 476 pure molecules and were asked to rate their smell along 21 different olfactory characteristics such as intensity, pleasantness, “garlicy,” sweet, “fishy,” “flowery” etc. to provide a perceptual rating. This was a purposefully diverse set of molecules, designed to cover a broad range of odors. Previous to this challenge, the limiting step in scent research was data availability; there were no models because there wasn’t enough data. The last comparable study was collected thirty years ago and contained twenty times less data.
Along with the human perceptual data set, challenge participants were given a comprehensive list of more than 4000 chemoinformatic features that describe the physical and chemical properties of the 476 molecules smelled by the individuals. Chemoinformatics is a new discipline at the intersection of computer science and chemistry designed to extract, at different scales, the most informative features from the structure of a molecule.
Over the course of several months, the teams developed algorithms and models to predict the correspondence between each volunteer’s perceptual rating and a given molecule. I remember the day when we saw the initial results – we just couldn’t believe how good the predictions were. Multiple modeling approaches performed astonishingly well. One could say participants of the DREAM Olfaction Prediction Challenge had a good “nose” for predicting smells! We were able to not only predict how the odor of a molecule would be perceived across the entire group of individuals, but also what a specific person would think the molecule smells like. Surprisingly, we found that regularized linear models models gave among the best predictions; meaning each subpart of a molecule creates an odor independently of the other parts. It’s a fascinating insight into the science of olfaction.
These results represent a giant leap towards solving a major challenge of understanding olfaction. Not only is this an important landmark in the neurobiology of olfaction, but the predictive models we describe may guide the design of new molecules for fields such as perfumery, flavor science, or other industrial applications.
The fuchsia color depicts the results of the best performing model, compared to the data for all 21 possible odor attributes (black outline) of the butyric acid molecule – a fatty acid found in Parmesan cheese. The model was effective in predicting the odor, particularly its intensity, pleasantness and musky qualities. It was less effective in predicting how woodsy it is; which is a more difficult odor to predict.
According to a report by Research and Markets, the global Flavors and Fragrances market is expected to grow at a rate of 7.0 percent over the next decade – reaching approximately $57.4 billion by 2025 (1). While flavors and fragrances only make up about 1 to 5 percent of the cost of a finished product, it is one of the key differentiators in product choice for consumers. We know that scent is the strongest sense tied to memory and emotion, with a direct line into the brain’s cortex (2). Pairing unique scents with product experiences builds deeper memories and connections with the brand, promoting both long-term loyalty and sales. Many of us can remember the smell of the laundry detergent our mothers used, even resulting in multiple generations of people purchasing the same product.
Extending beyond the scents of individual products, ambient smells in a shopping environment have also proven to influence consumer behavior. According to a study in the Journal of Marketing, it demonstrated that the “temperature” of scents in a store has a powerful effect on what, and how much, customers buy. Customers who smelled “warm” fragrances (like cinnamon) were more likely to purchase items they thought raised their personal status. Compared to those smelling “cool” scents, or not smelling any scent, those smelling “warm” scents bought more items overall (3).
Emotional ties to scent aren’t the only differentiating factor. The expertise required to become a Flavor Chemist (flavorist) is extensive – requiring undergraduate studies in food science or chemistry. Achieving membership into the Society of Flavor Chemists requires completing a seven year apprenticeship period and pass a review by their membership committee. Not limited to developing food flavors; the same approach is taken for creating new flavors and odors in consumer products such as cleaning and personal hygiene products. Let’s not even talk about oenologists, baristas and perfumers, who probably need the most exquisite sense of smell when rating wines, choosing coffee grains or developing new fragrances. By applying a clear predictive approach we could shorten the time it takes to create new or custom-tailored odors. Understanding scent at a molecular level could allow us to have a “scent-dial,” turning up or down a smell, or components of a smell, based on personal preference.
While more research is needed, especially in the area of mixtures of molecules, this research brings cognitive science to an age-old empirical domain.
PAGs play a vital role in the manufacturing of computer chips. They are also one of several classes of chemical compounds that have recently come under enhanced scrutiny from environmental regulators. Researchers have been racing to create more sustainable ones – but the traditional process of discovering new materials is too slow, too costly, and too risky. So IBM researchers have turned to AI for help – and created new PAGs much, much faster, paving the way to the era of Accelerated Discovery.