During the production of fermented foods such as kefir, Parmesan, and mountain cheese—as well as in protein powder manufacturing—bitter-tasting peptides can form. These compounds impair the taste and thus the acceptability of the final products, posing a challenge for food producers.
A research team led by the Leibniz Institute has developed an AI model that predicts which peptides will taste bitter. By analyzing the molecular structure of peptides, the model can identify bitter candidates before they are ever synthesized, saving time and resources in product development.
The model also opens the door to novel flavor design. Instead of simply removing bitter peptides, manufacturers could use the AI to adjust processing conditions or select enzyme treatments that minimize bitterness while preserving desirable flavor characteristics. This approach could help create healthier, tastier fermented foods and protein-rich products.
Because the source describes a single study, the article reflects only the team's reported findings. Independent validation and real-world application in industrial settings remain to be seen, but the approach represents a practical step toward data-driven flavor engineering.