Machine learning analysis of pre-culture effects on rate-limiting steps in volatile compound dynamics of Mead.

A novel two-step fermentation process was developed to enhance mead flavor quality. Headspace Solid-Phase Microextraction Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS) with three columns was used to analyze the volatile profiles of meads, along with sensory evaluation and machine learning. Compared to traditional mead (TM), our novel mead (NM) reduced off-flavor compounds by 37.6 %, with isoamyl alcohol decreasing 1.26-fold and ethyl laurate 2.09-fold. Meanwhile, aromatic compounds increased by 39.41 %, with isoamyl acetate rising 3.31-fold, ethyl caproate 2.79-fold, and phenylethyl alcohol 1.69-fold. Sensory evaluation revealed a significant reduction in bitterness (41.1 %) and irritation (42.5 %), while fruity, sweet, and pleasantly sour flavors increased by 27.4 %, 36.9 %, and 45.5 % for NM. Key aroma compounds (benzaldehyde, 2,3-butanediol, cedrol) were identified via recombination and omission experiments. Dynamic monitoring and machine learning identified key rate-limiting steps, including the oxidation of benzeneacetaldehyde (phenylethyl alcohol synthesis), isovaleraldehyde (isoamyl alcohol synthesis), and the conversion of octanoic acid to decanoic acid.