Using artificial intelligence to smell the roses

A pair of researchers at the University of California, Riverside, has used machine learning to understand what a chemical smells like — a research breakthrough with potential applications in the food flavor and fragrance industries.

We can now use artificial intelligence to predict how any chemical would smell to humans. It is possible to replace chemicals that are poisonous or harsh in, say, flavors, cosmetics, or household goods with cheaper, smoother, and healthier chemicals.

When any of their almost 400 odorant receptors, or ORs, are activated in the nose, humans detect odors. Each OR is triggered by a specific collection of chemicals; together, a vast chemical space can be detected by the large OR family. A central problem in olfaction is how receptors lead to specific properties or experiences of perceptuality.

The strength of learning machines is that they can analyze a large number of chemical characteristics and learn what makes a chemical smell like, say, a lemon or a rose or something else. Eventually, the machine learning algorithm will predict how a new chemical would smell, even if we do not know if it smells like a lemon or a rose at first.

It helps us to easily identify chemicals that have a different mix of smells. The technology will enable us to discover new chemicals that might replace old ones, for example, that are becoming scarce or very costly. It gives us a wide palette of compounds for every olfactory application that we can mix and match. For instance, you can now create a mosquito repellent that works on mosquitoes but smells good to humans.

The researchers first created a computer method for learning chemical properties that activate known human odorant receptors. They then screened about half a million compounds for new ligands for 34 odorant receptors — molecules that bind to receptors. He then concentrated on whether the algorithm that could estimate odorous receptor activity could also predict various odorant perceptual qualities.

The researchers demonstrated successfully the ORs behavior predicted 146 different experiences of chemicals. To their surprise, few were required to predict any of those expectations, rather than all ORs. Although they were unable to record sensory neuronal activity in humans, they further checked this in the fruit fly (Drosophila melanogaster) and found a similar result when predicting the attraction or aversion of the fly to specific odorants.

Many of the products available to customers make themselves appealing using volatile chemicals. Approximately 80 percent of what is called flavor in food is primarily derived from odors that influence smell. Cosmetic perfuming fragrances, cleaning materials, and other household items play a significant role in consumer behavior.

We now have a rare opportunity to detect ligands and new aromas and fragrances. We can intelligently design volatile chemicals that smell suitable for use using our computational approach and also predict ligands for the 34 human ORs.

Source: Joel Kowalewski, Anandasankar Ray. Predicting Human Olfactory Perception from Activities of Odorant Receptors.