Jerusalem, 8 November, 2026 (TPS-IL) — Israeli researchers have developed a new way to search for sugar substitutes that could offer the sweetness people want without the bitter aftertaste associated with some artificial sweeteners. Their approach uses a computer algorithm originally designed to identify potential drugs to find entirely new compounds that could eventually be used in food and beverages.
Unlike research focused on improving familiar sweeteners, the new approach could help identify entirely different families of sweet-tasting compounds, potentially opening new directions for the development of low-calorie food products.
A research team led by Dr. Shayma El-Atawneh and Prof. Amiram Goldblum at the Hebrew University of Jerusalem adapted a machine-learning system originally developed by Goldblum’s group for drug discovery, applying it to the search for potential sweetening compounds. The scientists stressed that the 92 compounds are computer-generated predictions, not confirmed sweeteners.
The study — published in the peer-reviewed journal Food Chemistry — addresses a longstanding challenge for food manufacturers. Artificial sweeteners such as aspartame, saccharin, and sucralose offer sweetness with fewer or no calories than sugar, but some leave bitter or metallic aftertastes. The researchers sought to identify molecules predicted to activate the body’s sweet-taste receptor without also triggering bitterness.
A New Approach to Sweetness
Goldblum’s group developed the machine-learning system, called Iterative Stochastic Elimination (ISE), for drug discovery. The researchers adapted it to identify chemical features associated with sweet- and bitter-tasting substances. The models examine 205 molecular properties, including the distribution of electrical charges and how molecules interact with oil and water.
The team applied the models to three chemical databases containing approximately 50.7 million molecules. It then combined the predictions with molecular docking, a computer simulation used to estimate how a compound might bind to a biological target.
The target was the human sweet-taste receptor, a protein complex made up of two components, T1R2 and T1R3. Structural information about the receptor, published in 2025, enabled the researchers to model how candidate molecules might bind to it.
The combined screening process narrowed the pool to 556 compounds, from which the researchers selected 92 candidates for further investigation. Forty-three had no more than one chiral center, a feature that can complicate chemical production and testing because molecules can exist in mirror-image forms. The remaining 49 were prioritized because the models predicted stronger sweetness and lower bitterness.
The researchers also found that 14 of the 92 candidates had previously appeared in patents describing compounds that interact with the sweet-taste receptor.
The scientists also identified an important weakness in the model: it incorrectly classified sucralose as unlikely to be sweet. They attributed the error to limitations in the chemical examples used to train the algorithm.
The next step is to test the candidates experimentally to determine whether they actually taste sweet, how intense their sweetness is, and whether they are safe for use in food products.
