Can Google smell? Why odor digitization could be a business opportunity

Artificial intelligence is entering the field of smells. Google researchers recently announced that they have trained an AI-powered neural network model to "map" the correlation between a molecule's structure and its smell, and it could spark a revolution in food. , perfume and health.

Let's say, for example, that your company manufactures insect repellents. Google researchers found that by feeding their neural network with data on the effectiveness of various molecules in repelling mosquitoes, the resulting model can go on to predict the mosquito repellency of almost any molecule. Humans are able to smell things because microscopic molecules are processed by receptors in your nose, which then send a message to your brain. The researchers found that more than a dozen molecules tested showed repellency at least as strong as DEET, the active ingredient in most insect repellents. These molecules could form the basis of a cheaper, longer lasting and safer spray.

Google researchers say smell is the most difficult sense to quantify in machine-readable data, and they've been working for several years to train their AI models to predict the smell of molecules by analyzing their structural composition. By feeding the network with data containing the composition of more than 5,000 molecules, coupled with multiple odor descriptors for each molecule, the researchers were able to create what they call a "master odor map." The map identifies the relationships between different scents by plotting thousands of data points, each representing the structure of an individual molecule analyzed by the neural network.

In practice, the map works much like a color wheel, with molecules of similar structure grouped together in the same way that similar colors like red and orange would be grouped together. The researchers found that nearby molecules tended to have the same odor descriptions.

In a test where trained panelists were asked to identify the odors of 400 different molecules using 55 descriptive labels, the neural network was able to accurately predict panel consensus responses far better than any what an individual panelist. The neural network was also able to assess the strength of particular smells.

Researchers say the AI-generated map allows them to predict the smell of billions of currently unknown molecules, which could have "broad applications in flavors and fragrances."