The structure of the molecule we inhale influences what we smell. No one can figure out how to do it. Scientists have figured out a few rules that govern how the nose and brain perceive airborne particles. The scent of garlic and certain ammonia-derived amines have become clear to us as the scent of garlic and certain ammonia-derived amines have become clear to us as the smell of garlic and certain ammonia-derived amine have become clear to us as the smell of garlic and certain ammonia

It turns out that the same molecule can have different smells. They both smell like almonds. Structural changes, even shifting the location of a double bond, can change a scent.

Artificial intelligence has been used to make sense of the chemistry. One team has trained a type of artificial intelligence called a graph neural network to predict what a compound will smell like to a person based on the chemical features of odor molecule. The researchers reported in a new draft paper that the computer model assessed new smells as reliably as humans.


It has learned something fundamental about how the world smells and how smell works, which was amazing to me.

A person's nose contains about 350 types of olfactoryreceptors, which can bind to a lot of airborne molecules. The brain eventually interprets a whiff of coffee, gasoline or perfume as a signal. Although scientists know how this process works in a broad sense, many of the details, such as how the system encodes these complex signals, still don't exist.

Vials with white caps, each with a black label of the smell sample it contains.
An “olfactory reference kit” of various known scents. Credit: Joel Mainland

Stuart Firestein, an olfactory neuroscientist at Columbia University, calls the model a "tour de force work of computational biology."

The model skips over the inscrutable workings of the nervous system in order to make direct connections between molecule and smell. Firestein describes it as a potentially useful tool to study the sense of smell and chemistry. The researchers hope that a more precise, numbers-based means to describe the odor universe will eventually bring this sense to the digital world.

In a future where computers can see, hear, and smell in the same way that humans can see, hear, and smell, I believe that datememe datememe can be commercially viable.


Computational modelling has been used to investigate olfaction. In a paper published in the same year, a model was created that could match the structure of a scent with a label. A map of odors was created by the model. Unlike a conventional paper map, which plots locations in two dimensions, the model placed odor molecule at "locations" based on the number of dimensions.

An illustration of the odor map (musk, lily, cabbage and grape)
An illustration of the odor map. Credit: Alexander B. Wiltschko

If the map held up to human perception, the team turned to the Monell Chemical Senses Center. It is difficult to define success in this area. The person says that. There is a panel of people who describe what it smells like.

Mainland and others identified a set of smells that had not been recorded. The researchers averaged the participants' assessments and compared them to the predictions of the model. For 53 percent of the molecule, the model came closer to the panel average than the individual panelist did.

You can sign up for Scientific American's newsletters.

The new model would not fare well in the real world, even though it may have been able to mimic human odor perception. Most smells are combinations. The team used perfumery data to train the new model and it skewed toward pleasant odors.

The model can still help those interested in the chemistry of smell, for example, guiding researchers who want to identify bad smells or test how changes in the structure of the brain affect perception. It could be used by fragrance chemists to identify new ingredients.


The model was used by the team to test a theory about the smell of a chemical. An animal's metabolism, the chemical processes that sustain its life, such as turning food into energy, could hold the explanation. They used their smell-map model to analyze the compounds they picked from the database. Molecules that are related to one another tend to smell the same even if they have different structures. Mainland was not a co-author of this paper but consulted with the team on the project. He says that they're not just building a model that fixes a problem. What is the underlying logic behind all of this?

The model may lead to the creation of new technology that can record or produce specific scents on demand. The work of the team is described as a step towards a complete map of human smell perception. The final version would be similar to the International Commission on Illumination's definition of a "color space". A professor of cognitive science at the University of Oxford who was not involved in the studies says that color space does not rely on words. She says that speakers of different languages have different ways of referring to the world. English speakers often refer to a smell as coffee or cinnamon. Jahai is an indigenous language that is spoken in Malaysia and parts of Thailand.

We don't know how this work will scale to other languages without empirical research. It is more difficult to distinguish between similar odors when asked to compare them, so researchers could have defined odors without labels. This approach proved less realistic according to Mainland. He thinks the map will be applicable elsewhere in the world because the model has learned something fundamental about the odor universe.

Even though it is possible to study human perception of odors without using words, researchers still don't have the ability to represent them in numbers. Researchers aim to describe smells with new precision by developing the olfactory equivalents of color space coordinates or hex codes.

The features of the brain that pay attention to vision and hearing have been learned by researchers. There are many things that need to be resolved for olfaction.


The identification of primary scent is one of the biggest challenges. To create the equivalent of digital imagery, in which smells are recorded and efficiently re-created, researchers need to identify a set of odor molecule that will reliably produce a range of smells when mixed.

Although people are working on it, it is very far out science fiction.