From a pool of 50.7 million chemical compounds, researchers in Jerusalem have picked out 92 that could one day help sweeten foods without the bitter edge found in some sugar substitutes. The candidates are not ready for a spoonful of tea: they are computer predictions that still need to be tested in the laboratory.
Dr. Shayma El-Atawneh and Professor Amiram Goldblum of the Hebrew University of Jerusalem built a computer-based search system to tackle a stubborn problem in sweetener research. People want alternatives to sugar, but familiar options such as aspartame, saccharin and sucralose can leave bitter or metallic aftertastes. Finding compounds that taste sweet while avoiding unwanted flavors is difficult when scientists must sift through huge numbers of possible molecules.
The team used a machine-learning method called Iterative Stochastic Elimination, or ISE. It learns patterns in the molecular properties of compounds already known to taste sweet or bitter, then uses those patterns to predict how other chemicals might taste. The researchers applied their models to three large chemical databases containing about 50.7 million molecules, filtering out candidates predicted to be bitter.
They then added molecular docking, a computer technique that estimates how a molecule might fit against a biological target. In this study, that target was the human sweet-taste receptor, made up of two protein parts called T1R2 and T1R3. A detailed structure of the receptor, published only last year, gave the researchers a newer map for examining how candidate molecules might bind to it.
The combined screening narrowed the vast starting pool to 556 potential sweeteners. The researchers then prioritized 92 candidates based on their predicted properties and suitability for future testing. The study, published in the journal Food Chemistry, also found that many of the candidates have structures unlike those of known sweeteners. That raises the possibility of exploring new kinds of sweet-tasting compounds, rather than only variations on familiar ingredients. Some candidates were also previously patented sweet-receptor modulators, offering an extra point of support for the search method.
The work matters because it could make the early stages of sweetener discovery faster and more focused. Instead of testing millions of compounds one by one, researchers could use the models to choose a smaller group for experiments. That may save time and effort for scientists and, eventually, help food and drink makers look for low-calorie options with a more pleasant taste.
But the 92 molecules are leads, not proven ingredients. Their sweetness, aftertaste, safety and stability have not been established by this computer search. Laboratory experiments must determine whether the predictions hold up, and further work would be needed before any candidate could become a product. For now, the study offers a promising way to narrow the search—and a reminder that artificial intelligence can help researchers find possibilities hidden in very large collections of chemicals.
