Scientists have learned to digitize complex odors with the help of artificial intelligence.
Modern.az reports that an international research group, including scientists from the Weizmann Institute, used machine learning to recognize complex odors and describe them mathematically. The results of the study were published in the scientific journal “PNAS”.
Scientists have stated that, unlike colors, there has been no universal mathematical system to describe odors until now. The reason for this is that most odors consist of dozens, even hundreds, of different molecules that interact with each other.
The researchers tried to create a system that could determine the similarity between different odors. For this purpose, a standard database consisting of 168 individual molecules and 731 odor mixtures was prepared within the framework of the “DREAM challenge” scientific competition.
26 scientific teams participated in the development of algorithms. They created models that predict how similar different odor pairs would be perceived by humans. After combining the most successful approaches, a model that predicts with high accuracy was obtained.
In addition, the algorithms described odors more accurately not only by the chemical properties of the molecules, but also by characteristics more familiar to humans, such as “fruity”, “sweet” or “woody”. Researchers believe that this result shows that the description of odors through language reflects the characteristics of the human brain in processing aromatic information.
Digitization of odors can be of practical importance in a number of fields. In medicine, this technology can be used to identify odor markers specific to certain diseases, including diabetes and liver diseases.
In the food industry, mathematical description of odors can help standardize the taste and quality of products, and in perfumery, it can accelerate the development of new perfumes and preserve unique compositions.
Researchers believe that this technology can reduce the need for a long-term trial and error process in the development of new aromatic products, replacing part of it with mathematical modeling.
