Israeli Scientists Discover Potential New Sugar Substitutes Using AI
Translated & summarized from NEWSru Israel by baba
Hebrew University scientists in Jerusalem used AI to screen 50 million molecules, identifying 92 potential new sugar substitutes. The AI algorithm predicted taste and binding to sweet receptors, significantly narrowing the search. While laboratory tests are still needed, the discovery could lead to novel low-calorie sweeteners with potentially fewer aftertastes.
The story in 5 lines · by baba
- Hebrew University scientists used AI to identify 92 potential new sugar substitutes from 50 million molecules.
- The AI algorithm predicted taste and binding to the human sweet taste receptor.
- The identified candidates have structures unlike current sweeteners, suggesting new classes of compounds.
- The research aims to address health issues linked to excess sugar consumption.
- Laboratory testing is required to confirm sweetness, safety, and food suitability.
Researchers at the Hebrew University of Jerusalem have utilized machine learning to identify 92 potential new sugar substitutes from a pool of 50 million molecules. The growing demand for low-calorie sugar alternatives stems from the health risks associated with excess sugar consumption, including obesity, type 2 diabetes, fatty liver disease, and cardiovascular issues. However, many existing artificial sweeteners like aspartame, saccharin, and sucralose can have unpleasant bitter or metallic aftertastes.
The team employed an AI algorithm called ISE, developed within Professor Amiram Goldblum's research group. This model was trained on known sweet and bitter substances to predict a molecule's taste based on its properties. The algorithm was then applied to three chemical databases containing approximately 50.7 million molecules.
The process began by filtering out compounds predicted to be bitter. Subsequently, the AI modeled how each molecule would bind to the human sweet taste receptor, specifically the T1R2 and T1R3 protein subunits, whose detailed structure was only recently published. This predictive modeling, combined with structural filters, narrowed the list down to 556 compounds.
Further selection reduced this to 92 candidate molecules. Notably, many of these candidates possess chemical structures dissimilar to current sweeteners, suggesting they may represent entirely new classes of sweet compounds. The inclusion of some already patented sweeteners among the candidates serves as indirect validation of the AI's effectiveness.
The researchers emphasize that these findings are currently computer predictions. The actual sweetness, safety, and suitability of these compounds for food applications require rigorous laboratory testing. Nevertheless, the AI's ability to significantly reduce the search space from tens of millions to a few dozen molecules drastically streamlines the experimental process.