In brief
Hebrew University researchers deployed an AI framework that screened about 50 million compounds and identified 92 candidates with desirable sweet taste and low bitterness, according to their study published in *Food Chemistry*. The models also predict how the compounds might interact with the human sweet‑taste receptor, aiming to reduce aftertastes and open new classes of sugar alternatives.
In a recent announcement on Sunday, researchers at Hebrew University in Jerusalem reported that a machine‑learning framework sifted through more than 50 million chemical molecules to spot 92 promising sugar‑substitute candidates, as detailed in a paper published in the journal Food Chemistry and communicated by Xinhua.
The team began with known sweet and bitter molecules to train predictive models that distinguish desirable from undesirable taste characteristics. These models were then applied to three chemical databases, collectively containing approximately 50.7 million molecules. The initial screening filtered out compounds predicted to be bitter, leaving a set of promising sweet candidates for further analysis.
Refining the Candidate List
After the first round of filtering, the researchers narrowed the list to 556 potential sweeteners. They then employed additional machine learning algorithms and molecular docking simulations to explore how each candidate might bind to the single human sweet‑taste receptor. Recent breakthroughs in the protein’s structural mapping, made public last year, provided the necessary detail for these binding predictions.
From the refined group, the researchers selected 92 compounds whose predicted taste profiles and receptor interactions were most favorable and considered suitable for future laboratory testing. Many of these candidates exhibit chemical structures distinct from existing sweeteners, suggesting the possibility of entirely new classes of sugar alternatives.
Implications for Sweetener Development
The study’s authors argue that artificial intelligence and molecular modeling can accelerate the discovery of new sweeteners while reducing unwanted aftertastes such as bitterness or metallic notes—a long‑standing challenge for the industry. By predicting taste profiles and receptor binding early in the development cycle, researchers hope to streamline the pipeline from discovery to market‑ready products.
According to the publication, the machine‑learning framework used by the Hebrew University team could be adapted to screen for other desirable food attributes, such as low glycemic index or enhanced nutritional properties. While the current work focuses on sweetness and bitterness, the authors note that the same approach could integrate additional sensory or functional criteria in future studies.
This breakthrough arrives amid a broader industry push toward low‑calorie, health‑oriented sweeteners. Companies and contract research organizations increasingly rely on AI to identify novel molecules that comply with regulatory standards and consumer expectations for clean‑label ingredients.
However, the study does not yet address regulatory approval pathways or large‑scale synthesis feasibility for the 92 candidates. The researchers plan to advance selected compounds into experimental validation, including cell‑based assays and animal studies, to confirm taste predictions and safety profiles.
In the wider context of food science, the Hebrew University work adds to a growing body of literature demonstrating AI’s role in accelerating ‘smart chemistry.’ Similar initiatives in pharmaceutical and agrochemical sectors have already leveraged machine learning for new drug discovery, crop traits, and sustainable chemical production.
While the technology remains in an early stage for dietary applications, stakeholders express cautious optimism. If the predicted sweeteners prove effective in real‑world testing, they may provide an alternative to high‑calorie sugar, addressing public health concerns about obesity and diabetes.
Nonetheless, the impact of this research will depend on subsequent experimental validation and the ability of academic and industrial partners to translate the findings into commercially viable products.
