Molecular Atlas of Key Food Odorants Reveals Mixture‐Level Organization and Enables Generative Aroma Design

Abstract

Aromas arise from complex combinations of odorants, yet how these mixtures encode stable and recognizable aroma identity remains unclear. Resolving this gap is key to both basic olfaction research and translational aroma design. Food provides a unique real-world system in which diverse mixtures produce well-defined aroma identities. Here we present KFO-Atlas, a molecular atlas of 896 key odorants curated from 2,282 food-derived aroma profiles. Analysis shows that every measured food aroma comprises at least three key odorants. Plant-derived food mixtures generally exhibit greater diversity in key odorant composition than animal-derived foods. Notably, in certain cases, distinct systems (e.g., plant- and animal-based foods) converge on a similar key odorant composition via shared reaction pathways. Building on these insights, we develop KFO-Gen, a generative AI model that produces category-targeted aroma formulations and validate its outputs by blinded human sensory evaluation. As a proof of principle, the model reconstructs meat-like aromas using exclusively plant-derived odorants, highlighting the potential of AI-guided aroma design for sustainable food innovation. KFO-Atlas and KFO-Gen establish a foundation for mixture-level studies of aroma, advancing both fundamental understanding and generative design.

The authors' abstract, as published at the source. Advanced Science, 2026 · DOI ↗