In brief
Hebrew University researchers screened 50 million compounds with AI, narrowing to 92 sweet‑tasting candidates that might be less bitter and could lead to new classes of sugar substitutes.
Researchers at Israel’s Hebrew University used a new machine‑learning framework to sift through 50 million chemical molecules and identify 92 promising low‑calorie sweeteners.
The team first built predictive models based on known sweet and bitter compounds, then applied those models to three large chemical databases containing roughly 50.7 million molecules. The AI flagged candidates that were predicted to taste sweet and not bitter.
After the initial filtering, the scientists used machine learning and molecular docking to simulate how each candidate might bind to the human sweet‑taste receptor, the protein whose detailed structure was published last year. This step helped refine the list to 556 potential sweeteners with favorable predicted properties.
From that pool, the researchers selected 92 compounds they judged to have the best balance of sweetness, minimal aftertaste, and suitable chemical structures for future testing. Many of the chosen molecules differ chemically from existing sweeteners, opening the possibility of entirely new classes of sugar alternatives.
These findings, published in the journal Food Chemistry, demonstrate how AI and molecular modeling can accelerate the discovery of sugar substitutes and reduce unwanted bitter or metallic aftertastes that have long challenged the industry.
