How artificial intelligence is deciphering the unknown biology of smell

Smell is one of the senses that arouses the most interest in the community. scientist because there are multiple aspects of its functioning that specialists have not yet managed to decipher. At least, not with the expected depth.

One of the biggest doubts concerns what The chemical structure of a molecule says almost nothing about its smell. .

As saved Nature , Two chemicals with very similar structures can smell very different .

And on the other hand, two very different chemical structures can produce an almost identical odor .

Added to this is that Most odors are mixtures of dozens or hundreds of aromatic molecules, which makes them more difficult to understand. in detail what is the link between the olfactory experience and the chemical structure.

There is also no clarity on how smells are related to each other .

Monell Chemical Senses Center neuroscientist Joel Mainland assured the aforementioned magazine that, compared to sight and hearing, olfactory parameters are less obvious .

Based on this premise, he pointed out that animals and humans have developed a decoding system that stands out for its great complexity .

For reference, for comparison, The human eye has two main types of receptors : cones and rods.

Instead, The smell has about 400 and it is not known how its signals combine to generate a specific perception.

Faced with this scenario, some teams believe that THE progress in molecular biology, data analysis and artificial intelligence (AI) can help answer the questions that captivated the community.

The eyes of many scientists are focused on deciphering the olfactory code .

It is assumed that understanding the biology behind this sense will allow better understand how animals use it to find food or a mate, and how smell is linked to factors such as memory, emotions or appetite .

There are also experts who are trying to develop devices capable of diagnosing diseases based on smells among other devices intended for areas such as health, insect repellent and even perfumes.

Sandeep Robert Datta, a neuroscientist at Harvard Medical School, summarized in Nature that “Smell is in fashion,” as experts in several fields see it in which its further study can contribute to different areas.

There, precisely, AI is positioning itself as a promising tool .

With these goals in mind, we have developed computer models that can link structures to odors .

Although early versions of these systems often relied on limited data sets or could only operate under prepared conditions, they have been observed. improvements in recent years .

For example, in 2020, a team presented a model with the ability to identify the most similar rose and violet scents that some of them with spicy spices like asafoetida.

Another of the projects that stands out the most is that of Osmo, a company led by researcher Alex Wiltschko and which aims to create a system capable of detecting, predicting or creating odors .

In August 2023, a group consisting of him, Mainland and other specialists presented a smell map using AI .

For that, They trained the model with thousands of descriptions of molecular structures and odor labels. which included terms like “floral” or “fleshy” to classify.

After, They compared their AI system to human noses .

For that, They trained 15 panelists So they could evaluate hundreds of aromas using 55 labels, including some like “tropical” and “smoky.”

When analyzing the results, which were subsequently published in the journal Science —, the authors saw that The AI ​​model showed strong results when predicting the odor of compounds .

And although the map he created is complex and has over 250 dimensions, was able to group the odors by type based on the data provided.

The mainland side stressed that The accuracy of the model was possible, in large part, thanks to its algorithm .

He thus referred to the fact that humans can, for example, calling a smell “fruity” but forgetting to say that they perceive it as “sweet” .

But the system can analyze different possibilities more comprehensively and patient.

One of the following objectives is which can predict the odors of compounds based on their components .

Along with this, They hope I can design new flavors which could have goals such as imitating a specific object or creating a more durable object.

However, Jane Parker, a flavour chemist at the University of Reading and co-author of the aforementioned study, said: Although the model can give “an idea of ​​what might work,” the work of scientists and specialists is essential. in related fields.

These are just some of the research which have been developed in this area.

And while these efforts have yielded promising results, There are key aspects to decipher about smell .

Biologist Linda Buck, who won the Nobel Prize in 2004 with her colleague Richard Axel for their discoveries about olfactory receptors, told Nature that “It would be nice to know how the perception of a particular odor is obtained” .

“How does this happen in the brain? Nobody knows. We still don’t have the techniques to know. “, said the scientist.