AI can infer a chemical’s odour from its structure

By examining the molecular structures of compounds, an artificial intelligence system may explain how they smell. Often, these descriptions resemble those of skilled human sniffers.

The system’s creators utilized it to create a list of smells, such as “fruity” or “grassy,” that are associated with hundreds of chemical compounds. This odorous guidebook may aid researchers in the development of new synthetic fragrances as well as providing insights into how the human brain interprets smell.

A whiff of a memory

Smells are the only sort of sensory information that travels directly from the sensory organ — in this case, the nose — to the brain’s memory and emotional centres; all other types of sensory input pass through other brain regions first. This straightforward explanation explains why odours can elicit specific, vivid memories.

According to neurobiologist Alexander Wiltschko, smell is unique. His Cambridge, Massachusetts-based startup, Osmo, is a spin-off from Google Research that aims to create novel smelly molecules, or odorants.

Wiltschko and his team at Osmo created a form of artificial intelligence (AI) system known as a neural network that can assign one or more of 55 descriptive terms, such as fishy or winey, to an odorant in order to investigate the relationship between a chemical’s structure and its scent. The group gave the AI instructions to characterize the fragrance of over 5,000 odorants. To understand the connection between chemical structure and scent, the AI also examined each odorant’s chemical composition.

Around 250 associations between particular chemical structure patterns and a particular odour were found by the system. In order to help the AI forecast the scent of a novel chemical, the researchers aggregated these associations into a particular odour map (POM).

15 volunteers were trained to associate particular odours with the same set of descriptive phrases used by the AI so that the POM could be tested against human noses. The writers then gathered a large number of odorants that are artificial yet are nonetheless recognisable enough for individuals to describe. then asked 323 of the human participants to describe them, and then asked the AI to anticipate the scent of each new molecule based on its chemical makeup. The AI’s prediction frequently came closer than any individual’s approximation to the average human answer.

The nose’s knowledge

Stuart Firestein, a neurologist at New York City’s Columbia University, calls it a wonderful advancement in machine learning. The food and cleaning product industries, for instance, might benefit from the POM, according to him.

Firestein, however, notes that the POM doesn’t provide much information regarding the biology underlying the human sense of smell, such as how various compounds interact with the roughly 350 odour receptors in the human nose. He asserts that although they have the chemical side and the brain side, we are still in the dark regarding the middle.

The use of language in the paper to connect structures with individual odours is praised by Pablo Meyer, a systems biologist at the IBM Centre for Computational Health in Yorktown Heights, New York. He does not, however, think that the “correct” way to describe a smell is the average of what the humans said. According to him, smell is a subjective sense. There isn’t, in his opinion, a right way to perceive something.

The next stage, according to Wiltschko, is to discover how odorants interact and compete with one another to produce an aroma that the human brain perceives as being wholly distinct from the aromas of each of the individual odorants. This will be highly challenging, according to Meyer and Firestein: combining merely 100 molecules in various combinations of 10 results in 17 trillion variants, and the number of conceivable combinations quickly rises to a level that is simply too great for a computer to process.

But according to Firestein, that is the way people genuinely smell. A single fragrance, like coffee, can have hundreds of odorant compounds in it. The next step is to anticipate how the mixture will smell, according to Wiltschko.

Source link