Predicting Flavor Partitioning Using Artificial Intelligence Methods Focusing on Milk and Milk Alternatives

Key Facts

Full Project name (orignial):Prognose der Aromastoffverteilung mit Verfahren der künstlichen Intelligenz am Beispiel von Milch und Milchalternativen
Project number:01IF249010N
Funding guideline:Industrielle Gemeinschaftsforschung
Budget:578,475 €
Time period:April 2026 to September 2028
Funding:FEI (Forschungskreis der Ernährungsindustrie e.V.)
Project summary
Project manager:

Prof. Dr. Yanyan Zhang (research facility 1)

Prof. Dr. Christian Krupitzer (research facility 2)

Prof. Dr. Christine Borsum (research facility 3)

Person in charge:

Ruotong Nie und Caroline Pottgießer (research facility 1)

Dana Jox, M.Sc. (research facility 2)

Antonia Kunz (research facility 3)

Project description

Current trends in the food industry are increasingly focused on plant-based alternatives and reducing sugar, salt, and fat content, driven by health consciousness and sustainability goals. However, for these changes to lead to a long-term shift in dietary habits, plant-based alternatives must meet consumers’ sensory expectations. The perception of a food’s aroma—such as that of milk—depends on various parameters related to both the aroma compounds and the food matrix. It is particularly important to understand the interactions between representative aroma compounds and the composition of foods (e.g., fat and protein), as these significantly influence the release of aroma compounds and, consequently, the sensory perception of the food. The distribution of flavor compounds between the matrix and the headspace above the food is quantified by the Kmg value, which reflects the equilibrium in a sealed product. Flavor compounds interact noncovalently with food components such as fats and proteins. Changes in the type or content of fats and proteins can therefore influence the binding of flavor compounds and their release into the headspace. The Kmg value can be used to reflect the extent to which flavor compounds are released from food matrices of varying composition into the headspace. This project aims to better understand the effects of changes in matrix composition on the binding and release of flavor compounds. First, the distribution of flavor compounds in well-studied milk matrices will be modeled, followed by a machine learning (ML) analysis to identify the key factors influencing flavor release and to facilitate the adaptation of the model to plant-based systems. To this end, an ML method will be developed that can predict the Kmg value of flavor compounds in vegan milk alternatives with varying compositions. The research findings are significant for established dairy companies, as they enable these companies to meet the growing demand for vegan alternatives with shorter development cycles and to prepare for—or actively shape—upcoming changes in food production. A deeper understanding of flavor distribution in food matrices can support efficient reformulation, reduce undesirable off-notes (e.g., green or bean-like flavors in plant-based dairy products), and empower SMEs to meet the demand for high-quality vegan products. The findings would primarily benefit SMEs and the very active startup scene, as the ML models generated in the project can be integrated into easy-to-use open-source software for analysis.