In brief
Hebrew University scientists developed a machine‑learning framework that screened over 50 million compounds, reducing the field to 556 potential sweeteners and ultimately 92 candidates with favorable taste predictions and binding to the human sweet‑taste receptor. Many of the selected molecules differ chemically from existing sweeteners, suggesting new classes of sugar substitutes.
In a study released November 8, Hebrew University researchers announced they used an advanced machine‑learning framework to sift through more than 50 million chemical molecules, drilling down to 92 promising sugar‑substitute candidates. The work, published in the journal Food Chemistry, demonstrates how artificial intelligence and molecular modelling can accelerate the discovery of new sweeteners while potentially reducing unwanted bitter or metallic aftertastes that challenge the industry.
The researchers began by compiling predictive models that distinguish sweet from bitter taste profiles, using known sweet- and bitter-tasting compounds as training data. These models were then applied to three chemical databases containing approximately 50.7 million molecules, filtering out compounds predicted to be bitter and highlighting candidates likely to elicit sweet sensations.
After the initial screening, the team employed machine‑learning techniques and molecular docking to simulate how the candidate compounds might bind to the single human sweet‑taste receptor. The protein’s detailed structure, published last year, enabled the researchers to assess the binding affinity and potential efficacy of each molecule.
Following further refinement, the list was narrowed to 556 potential sweeteners. A final selection process, based on favorable predicted properties and suitability for future testing, resulted in 92 molecules. Many of these selected compounds have chemical structures distinct from those of existing sweeteners, potentially opening a door to entirely new classes of sugar alternatives.
"Israeli researchers have developed a new machine-learning framework that identified 92 candidates for future sugar substitutes from more than 50 million chemical compounds, the Hebrew University of Jerusalem said in a statement on Sunday," the universities statement read.
