Can the taste of a food be predicted from its ingredient list? In a new study, researchers including UC Davis professor Ilias Tagkopoulos explored this question by combining food science, artificial intelligence, and engineering-inspired modeling approaches. The team treated recipes as complex systems and developed computational methods to predict key taste dimensions—including sweetness, sourness, bitterness, umami, and saltiness—from ingredient-level data. Their findings demonstrate that ingredient composition alone does not fully explain how foods are ultimately perceived, highlighting the critical role of processing-related chemical transformations such as caramelization, Maillard reactions, and other mechanisms that shape flavor.
The research introduces an interpretable hybrid modeling framework that combines foundational scientific principles with data-driven AI approaches to improve taste prediction accuracy while maintaining transparency. Beyond prediction, the team also demonstrated how these models can be used to reverse-engineer ingredient formulations that achieve targeted taste profiles. The work reflects the type of translational food innovation advanced by Ilias Tagkopoulos and collaborators at UC Davis, helping build tools that could support the design of healthier, more sustainable, and consumer-acceptable foods through evidence-based computational approaches. While additional validation is needed across broader food categories and real-world applications, the study highlights the potential of AI-enabled food design to accelerate innovation across the food and health ecosystem.
