A pair of researchers at the University of California, Riverside, have used machine learning to understand what chemical odors are – a research breakthrough with potential applications in the food flavor and aroma industries.
Anandasankar Ray, professor of molecular, cell and systems biology and senior author of the study appearing in Ices, said, “We can now use artificial intelligence to predict how a chemical is going to smell a human being.” “Chemicals that are toxic or harsh, say, flavor, cosmetics, or household products, can be replaced with natural, soft and safe chemicals.”
When some of the nearly 400 olfactory receptors, or ORs, are activated in the nose, humans are understandable. Each is activated by a unique set of chemicals; Together, large or family can detect a huge chemical location. A major question in olfactory is how receptors contribute to different perceptual properties or perceptions.
“We tried to model human olfactory perceptions using chemical informatics and machine learning,” Ray said. “The power of machine learning is that it is capable of evaluating a large number of chemical characteristics and knowing what a chemical smell is, for example, lemon or rose or anything else.
Machine learning algorithms can eventually predict how a new chemical will smell, even though we may not initially know it smells like lemon or rose. ”
According to Ray, a new way of scientifically prioritizing the smell of chemicals is to get an idea of what kind of chemicals can be used in the food, flavor and aroma industries.
“This allows us to rapidly find chemicals that have a novel combination of odors,” he said. “Technology can help us discover new chemicals that can replace existing ones that are becoming scarce, for example, or which are very expensive.
This gives us a huge palette of compounds that we can mix and match for any olfactory application. For example, you can now make a mosquito repellent that works on mosquitoes but is enjoyable for humans. ”
Researchers first developed a method for computers to learn the chemical characteristics that activate human odor odor receptors. They then screened about half a million compounds for new ligands – molecules that bind to receptors – to 34 olfactory receptors. Next, they focused on whether the algorithm could predict olfactory receptor activity that could predict the diverse perceptual properties of olfactors.
“Computers can help us better understand human perceptual coding, which appears to be based on a combination of differentially active ORs,” said Joel Kovelvsky, a neuroscience graduate program working with Ray A student in and the first author of a research paper.
“We used hundreds of chemicals that human volunteers had previously evaluated, selected ORs that best predicted a portion of the chemicals, and tested whether these ORs were predictive of new chemicals as well.”
Ray and Kowalevsky successfully predicted 146 different perceptions of the chemicals the activity of ORS. To their surprise, some rather than all ORs were required to predict these assumptions. As they could not record activity from sensory neurons in humans, they further tested this in the fruit fly (Drosophila melanogaster) and observed similar results when predicting fly attraction or progression to different sulfur.
Ray pointed out that many items available to consumers use volatile chemicals to make themselves attractive. Approximately 80% of what is considered taste in food actually stems from odors that affect odor. Fragrances play an important role in consumer behavior for scented cosmetics, cleaning products, and other household items.
“Our digital approach using machine learning can open up many opportunities in the food, flavor and aroma industries,” he said. “We now have an unprecedented ability to discover ligands and new flavors and fragrances. Using our computational approach, we can intelligently design volatile chemicals that smell desirable for use and also predict ligands for 34 human ORs. ”
Odor perception in humans results from the activation of odor receptors (ORs) in the nose. However, ORs associated with specific olfactory perceptions are unknown, as opposed to sight or taste where receptors are associated with perception of different colors and tastes.
The large family of ORs (~ 400) and several receptors activated by an olfactory present serious challenges. Here, we use for the first time machine learning ~ 0.5 million compounds for the new ligand and identification of rich structural motifs for 34 human ORGs.
We next demonstrate that ORS activity successfully predicts many of the 146 different perceptual properties of chemicals. Although chemical characteristics have been used to model odor odor, we point out that biologically relevant OR activity is often preferable.