In brief
A machine‑learning framework developed at the Hebrew University of Jerusalem sifted through 50.7 million molecules and yielded 92 sugar‑substitute candidates with predicted sweet taste and minimal bitterness, according to Xinhua. The study was published in *Food Chemistry* and marks a potential step toward new low‑calorie sweeteners, though practical efficacy remains to be tested.
Israeli researchers at the Hebrew University of Jerusalem announced on Sunday that their new machine‑learning framework identified 92 promising sugar‑alternative candidates after screening 50.7 million chemical molecules (Xinhua). The team used predictive models trained on known sweet and bitter compounds to distinguish desirable taste profiles, then applied these models to three large databases to filter out bitter candidates (Xinhua).
After an initial screening that produced 556 potential sweeteners, the researchers refined the list with machine‑learning and molecular docking to simulate binding to the human sweet‑taste receptor, narrowing the final list to 92 molecules (Xinhua). Many of the selected compounds have structures different from existing sweeteners, raising the possibility of entirely new classes of sugar alternatives (Xinhua).
The findings were published in the journal Food Chemistry and show how artificial intelligence could accelerate discovery of new sweeteners that avoid the unwanted bitter or metallic aftertastes that have long challenged the industry (Xinhua). The authors note that the next steps will involve testing the candidates in the laboratory and eventually in food products, although those practical evaluations are still unconfirmed.
An unrelated press release about NNB Nutrition’s presentation of next‑generation sweeteners at SupplySide Global 2026 could not be retrieved, and a related science article notes the broader challenge of reproducing sugar’s sweetness without lingering bitterness (Earth.com).
