Abstract: Non-targeted mass spectrometry analysis based on ultra high performance liquid chromatography-quadrupole-time of flight-tandem mass spectrometry (UPLC-Q-TOF-MS) combined with machine learning can predict whether beer is infected by spoilage bacteria. The metabonomics was used to analyze beer spoilage bacteria such as Lactobacillus brevis, Bacillus, Lactobacillus parabuchneri, Lactobacillus plantarum and Staphylococcus saprophyticus during beer fermentation. The metabonomics data was then analyzed by the multivariate analysis method of random forest. The first 30% of the important characteristic substances of the model were selected for pathway enrichment analysis. Finally, it was found that in the initial stage of fermentation, the contents of choline, choline glycerophosphate, triethanolamine, phosphatidylethanolamine and 2-(3-carboxylpropionyl)-6-hydroxy-cyclohexa-2,4-dicarboxylic acid (SHCHC) in spoiled beer samples were higher than those in normal beer samples. The contents of choline in spoiled beer with 36-48 h fermentation were 50% - 130% higher than those in normal beer samples; The content of glycerophosphate choline in spoiled beer with 24 h fermentation was 10% - 30% lower than that in normal beer. As the fermentation goes on, the content of glycerophosphate choline in spoiled beer gradually increased to more than 9% higher than that in normal beer; The content of triethanolamine in normal beer with 24 h fermentation was the highest, which was 2-3 times higher than that in spoiled beer; Phosphatidylethanolamine in spoiled beer was 6% - 13% higher than that in normal beer with 96 h fermentation; SHCHC was also only detected in spoiled beer with 36 h fermentation. These substances could be used as biomarkers of beer spoilage.

Key words: machine learning, beer spoilage bacteria, metabonomics