【作者】 许江；

【导师】 毛健； 周新虎；

【作者基本信息】 江南大学 ， 生物技术与工程（专业学位）， 2025， 硕士

【摘要】 酒药是传统黄酒生产中一种重要的活性发酵剂,由多种微生物群落组成,包括酵母、霉菌和乳酸菌及其代谢产物。酒药中的微生物群落通过复杂的代谢活动将淀粉和蛋白质转化为酒精、有机酸、氨基酸和其他风味化合物,这些过程赋予了黄酒独特的风味和品质,使其在黄酒生产中不可或缺。其微生物群落的动态变化直接影响黄酒的风味品质和发酵效率。然而,传统的酒药发酵通常在开放环境中进行,这使得其微生物群落结构极易受到外部因素的影响。在这些微生物中,乳酸菌的作用尤为关键,乳酸菌不仅产生乳酸,降低发酵环境的p H值并抑制有害细菌的生长,还通过其代谢活动产生多种芳香化合物。在微生物生态学中,目前检测微生物群落中特定物种的方法主要依赖于传统培养技术和分子生物学方法。这些方法耗时周期长无法实时监测酒药发酵中乳酸菌群落结构变化,因此,开发快速、准确的微生物监测技术对酒药发酵具有重要意义。主要研究内容如下:（1）通过高通量测序技术,对中国不同地区23个酒药样品的微生物群落进行解析,鉴定优势菌群及其相对丰度,明确乳酸菌在发酵过程中的核心作用。筛选6种关键乳酸菌戊糖片球菌、融合魏斯氏菌、乳酸片球菌、乳酸乳球菌、肠膜明串珠菌、类肠膜魏斯氏菌,作为发酵过程监测的指示菌群。建立产乳酸微生物分选体系,基于p H荧光探针的微流控分选模型,实现乳酸菌的高效分离与纯化。通过对菌株序列和16S r RNA扩增子测序序列比对,得到目标乳酸菌。（2）采集6种乳酸菌的10,000条单细胞拉曼光谱,对光谱进行了峰强度和信噪比的评估,以去除异常值和校正宇宙射线的影响。随后,使用背景校正技术消除了光谱中的背景噪声,进一步提高光谱数据的质量。采用了自适应迭代重加权最小二乘法（air PLS）进行基线校正。使用Savitzky-Golay平滑算法对光谱数据进行平滑处理,以减少噪声并提高信噪比。最后,对光谱数据进行了最小-最大归一化处理,以消除不同样本之间的强度差异,使数据更具可比性。（3）采用用四种特征选择算法Laplacian、MICI、RF和Relief F对酒药中6种乳酸菌的单细胞拉曼光谱指纹区数据400 cm^-1-1800 cm^-1进行了系统分析,筛选出64个关键特征变量,主要位于3个波段范围:400-680 cm^-1、680-960 cm^-1和1240-1520 cm^-1,其中1240-1520 cm^-1区域贡献度最高。计算三个区域内乳酸菌之间单细胞拉曼光谱数据主要峰位强度后进行显著性分析,基于显著性结果得出对光谱区分贡献最大的波段分别为538 cm^-1-634 cm^-1、1320 cm^-1-1334 cm^-1和820 cm^-1-837 cm^-1。（4）根据6种分类模型的评估指标,Res Net-34模型对6种乳酸菌单细胞拉曼光谱数据的快速区分能力最好,鉴定准确率为98.3%,通过合成菌群验证ResNet-34模型,与实际样本中预测误差<2%,实际发酵监测结果与测序数据无显著差异（p>0.05）。基于PLS-R算法构建的发酵时间预测模型R²达0.927,RMSE为0.00345,可准确预测酒药发酵时间。更多还原

【Abstract】 Jiuyao is an important active fermentation starter in traditional Huangjiu production,composed of diverse microbial communities including yeasts,molds,lactic acid bacteria,and their metabolites.The microbial communities in Jiuyao convert starch and proteins into alcohol,organic acids,amino acids,and other flavor compounds through complex metabolic activities,which impart unique flavors and qualities to the wine,making it indispensable in yellow rice wine production.The dynamic changes in these microbial communities directly affect the flavor profile and fermentation efficiency of the wine.However,traditional Jiuyao fermentation is typically conducted in open environments,making the microbial community structure highly susceptible to external factors.Among these microorganisms,lactic acid bacteria play a particularly critical role,not only producing lactic acid to lower the p H of the fermentation environment and inhibit the growth of harmful bacteria but also generating various aromatic compounds through their metabolic activities.In microbial ecology,current methods for detecting specific species within microbial communities primarily rely on traditional cultivation techniques and molecular biology methods.These methods are time-consuming and unable to monitor real-time changes in the lactic acid bacterial community structure during Jiuyao fermentation.Therefore,developing rapid and accurate microbial monitoring technologies is of great significance for Jiuyao fermentation.The main research contents are as follows:（1）The microbial communities of 23 Jiuyao samples from different regions of China were analyzed using high-throughput sequencing technology to identify dominant bacterial groups and their relative abundances,clarifying the core role of lactic acid bacteria in the fermentation process.Six key lactic acid bacteria—Pediococcus pentosaceus,Weissella confusa,Pediococcus acidilactici,Lactococcus lactis,Leuconostoc mesenteroides,and Weissella paramesenteroides—were selected as indicator strains for fermentation monitoring.A lactic acid-producing microbial sorting system was established based on a p H-sensitive fluorescent probe microfluidic sorting model to achieve efficient isolation and purification of lactic acid bacteria.Target lactic acid bacteria were obtained by comparing strain sequences with 16S r RNA amplicon sequencing sequences.（2）A total of 10,000 single-cell Raman spectra were collected from the six strains of lactic acid bacteria.The spectra were evaluated for peak intensity and signal-to-noise ratio to remove outliers and correct for cosmic ray interference.Background correction techniques were then applied to eliminate background noise and further improve data quality.Adaptive iteratively reweighted least squares（air PLS）was used for baseline correction.The Savitzky-Golay smoothing algorithm was employed to reduce noise and enhance the signal-to-noise ratio.Finally,min-max normalization was performed to eliminate intensity differences between samples,ensuring comparability of the data.（3）Four feature selection algorithms—Laplacian,MICI,RF,and Relief F—were systematically applied to analyze the single-cell Raman spectral fingerprint region(400-1800cm^-1)of the six lactic acid bacteria strains in Jiuyao.Sixty-four key feature variables were identified,primarily located in three spectral regions:400-680 cm^-1,680-960 cm^-1,and 1240-1520 cm^-1,with the 1240-1520 cm^-1region contributing the most.The main peak intensities of the single-cell Raman spectra of lactic acid bacteria within these three regions were calculated,followed by significance analysis.Based on the results,the spectral regions contributing most to differentiation were identified as 538-634 cm^-1,1320-1334 cm^-1,and 820-837 cm^-1.（4）According to the evaluation metrics of the six classification models,the Res Net-34model demonstrated the best performance in rapidly distinguishing the single-cell Raman spectra of the six lactic acid bacteria strains,achieving an identification accuracy of 98.3%.The Res Net-34 model was validated using synthetic microbial communities,showing a prediction error of<2%in actual samples,with no significant difference（p>0.05）between the fermentation monitoring results and sequencing data.A fermentation time prediction model based on the PLS-R algorithm achieved an R²of 0.927 and an RMSE of 0.00345,enabling accurate prediction of Jiuyao fermentation time.更多还原