Predicting the low-level and extremely low-threshold compounds in Baijiu: uniform manifold approximation and projection

Food Chem X. 2025 Jun 10:29:102645. doi: 10.1016/j.fochx.2025.102645. eCollection 2025 Jul.

Abstract

In the flavor analysis of Baijiu, the identification of compounds exhibiting high flavor dilution factors but displaying minimal or undetectable response signals in analytical measurements, remains a significant challenge in current research. To clearly elucidate the correlation between the structure of odorants and their odors in Baijiu, a "compound-aroma" database consisting of 646 compounds and their known associated odors (70 aroma descriptors) has been compiled. Each individual compound was coded with 1024-bit molecular fingerprints and analyzed by K-means and uniform manifold approximation and projection (UMAP). Moreover, by calculating the presence or the absence of odor molecular substructures, associations for both odor notes and chemical groups of the aroma compounds in Baijiu were revealed, and an indicative model diagram was constructed. Finally, a strong roasted odor of Baijiu, which presented at the retention index of 1324-1334 with no distinct response signal, has been successfully identified as 2,5-dimethylpyrazine or 2,6-dimethylpyrazine by using the aforementioned model.

Keywords: Baijiu; Compounds prediction; Molecular fingerprints; Structure-odor correlations; UMAP.