In brief
Israeli scientists at Hebrew University used artificial intelligence to screen 50.7 million chemical compounds and pinpoint 92 molecules that could serve as next‑generation sugar alternatives. The study, published in *Food Chemistry*, combined predictive taste models with molecular docking to the human sweet‑taste receptor, aiming to reduce bitter aftertastes that have long plagued new sweeteners.
Jerusalem, Nov. 8 – Israeli researchers at Hebrew University announced on Sunday that a machine‑learning framework has sifted through more than 50 million chemical molecules to identify 92 promising sugar‑substitute candidates. The finding, published in the journal Food Chemistry, illustrates how artificial intelligence can accelerate the discovery of new low‑calorie sweeteners while addressing the persistent problem of bitter or metallic aftertastes.
According to a university statement, the team began with known sweet- and bitter-tasting compounds to train predictive models that could distinguish desirable sweetness from undesirable bitterness. After validating these models, the researchers applied them to three chemical databases, totaling approximately 50.7 million molecules. The AI sifted through the data, automatically scoring each compound for its predicted sweetness and filtering out those likely to taste bitter.
From the initial list, the researchers narrowed the candidates down to 556 molecules that exhibited favorable predicted taste profiles and were deemed suitable for further study. They then employed machine learning in conjunction with molecular docking techniques to simulate how each candidate might bind to the human sweet‑taste receptor, a protein whose detailed structure was released last year.
By integrating the predictive taste model with receptor‑binding simulations, the team was able to assess each compound’s potential efficacy as a sweetener before any laboratory testing. This dual‑layer approach—taste prediction followed by receptor compatibility—helped concentrate the top 92 molecules that exhibit both sweet taste and promising interaction with the receptor.
Notably, many of the selected compounds boast chemical structures that differ from those of existing sweeteners, opening the possibility of entirely new classes of sugar alternatives. The researchers suggested that this structural novelty could lead to sweeteners with unique sensory profiles and lower caloric content.
The AI Screening Process Explained
The study’s methodology rests on two pillars: predictive modeling of taste attributes and molecular docking to the sweet‑taste receptor. Predictive models were built using datasets of molecules with established sweet or bitter properties. The models output probability scores that indicate whether a new molecule will be perceived as sweet or bitter by human taste buds.
Once a compound’s predicted sweetness exceeded a set threshold, the team fed the molecule into a docking simulation. This simulation, powered by machine learning, estimates how strongly the molecule binds to the receptor’s active site. Molecules that bind effectively are more likely to trigger the sweet‑taste signaling cascade in humans.
By combining these two steps, the researchers effectively reduced the experimental burden. Instead of testing thousands of compounds in the lab, they focused resources on 92 candidates that exhibited the best combination of predicted sweetness and receptor affinity.
Implications for the Sweetener Industry
Food and beverage manufacturers have long struggled to replace sugar with low‑calorie alternatives that deliver both taste and mouthfeel. Bitter aftertastes and metallic notes often deter consumers, limiting the adoption of new sweeteners. The Hebrew University team’s AI-driven pipeline offers a systematic route to discover compounds that may overcome these sensory hurdles.
Industry experts note that the methodological framework could shorten the development timeline for next‑generation sweeteners. By eliminating compounds with unfavorable taste profiles early, companies may reduce costs associated with late‑stage formulation failures.
Furthermore, the discovery of structurally distinct molecules hints at potential breakthroughs in nutritional safety. Novel sweeteners may exhibit lower glycemic impact and improved metabolic profiles, aligning with the growing demand for healthier food options.
While the research represents a significant technical advance, the practical application of the 92 candidates will require further laboratory validation. Confirmatory taste panels, toxicity assessments, and regulatory reviews are essential next steps before any of the molecules can reach the market.
In summary, the Hebrew University study demonstrates the power of AI to streamline sweetener discovery, reduce unwanted aftertastes, and potentially introduce new classes of low‑calorie sugar substitutes. As the technology matures, it may reshape how the food industry approaches sweetener development.
