In brief
Hebrew University scientists announced a data‑driven process that sifted 50.7 million chemical compounds to find 92 promising sugar‑alternatives. The AI models evaluate sweetness, bitterness, and binding to human sweet‑taste receptors, promising new low‑calorie options for food manufacturers. The discovery could herald a new class of sweeteners with fewer aftertaste issues.
In Jerusalem, a team of Israeli scientists unveiled a new machine‑learning framework that sifted through more than 50 million chemical molecules to identify 92 promising candidates as future sugar substitutes, according to a statement released by Hebrew University on Sunday, November 8, 2026.
The breakthrough, detailed in a recent issue of the journal Food Chemistry, employs artificial intelligence to predict sweet and bitter taste profiles and simulate how potential molecules would bind to the human sweet‑taste receptor. By doing so, the researchers aimed to accelerate discovery of sweeteners that are low‑calorie and free from the unwanted metallic or bitter aftertastes that have plagued the industry for years.
The team began by assembling a training set of known sweet and bitter compounds, from which they built predictive models that could distinguish desirable from undesirable taste characteristics. Using these models, they screened three chemical databases that collectively contain approximately 50.7 million molecules, flagging those that both tasted sweet and avoided bitter markers. The initial pass yielded a broad list of promising sweet candidates, after which the scientists applied machine‑learning techniques alongside molecular docking simulations to forecast how each compound might interact with the human sweet‑taste receptor’s protein structure—a detailed map that became available last year.
After refining the data, the researchers trimmed the list to 556 molecules with favorable predicted properties and feasibility for future testing. From that pool, 92 candidates were selected as the most promising. Many of these compounds exhibit chemical frameworks that differ markedly from existing sweeteners, suggesting they could open a new class of sugar alternatives rather than merely tweaking current formulas.
For food manufacturers, the development could translate into a richer menu of low‑calorie sweetening options. The AI’s ability to predict taste outcomes before a single compound is synthesized means that developers could focus resources on the most promising leads, cutting time and cost typically associated with sweetener research.
Although the current study is computational, the research team plans to send the top candidates into laboratory testing to confirm sweetness, safety, and regulatory compliance. If successful, these new molecules may soon move from the lab to the grocery aisle, offering consumers a broader range of low‑calorie sweetener choices with fewer aftertaste problems.
