单细胞拉曼光谱(SCRS)技术具有快速、 灵敏和无标记的优势, 可以从单细胞水平上研究细胞结构, 本文为实时监测单细胞微生物生长代谢变化, 提出了基于谱聚类和SCRS的细胞生长检测方法, 并采集600个同步培养的发酵工程菌-大肠杆菌SCRS数据作为实验数据, 采集300个发酵益生菌-枯草芽孢杆菌SCRS数据验证方法适用性。 首先, 对同步培养的菌落测量OD600生长曲线作为微生物群体水平上生长时期标签; 其次, 应用t-SNE对群体细胞SCRS数据进行可视化分析, 指导谱聚类对高维SCRS数据聚类分析, 并应用轮廓系数和CH index评估最佳聚类簇, 赋予每个SCRS数据簇标签; 最后, 应用三次样条插值拟合统计SCRS数据簇标签和生长时期标签交集, 精准识别群体中共存的生长时期异质数据, 实现对单细胞微生物生长时期精准鉴定。 结果表明, 基于谱聚类与SCRS的细胞生长分析方法根据同步培养的群体细胞生长曲线, 设置2维嵌入空间维度和基于最近邻的谱聚类相似度计算方法, 有效检测三个生长时期最佳聚类簇中9%和4.3%异质数据。 提出的无监督检测单细胞生长的方法, 借助谱聚类无需标记就可以直接根据SCRS数据特征进行建模, 并能够对任意形状的高维SCRS数据聚类且快速收敛的优势, 实现了对两种发酵工程菌和发酵益生菌细胞滞后期、 对数期和稳定期的精准识别, 真正意义上实现从单细胞水平上检测细胞生长, 为发酵工程提供更加精准、 实时的调控指导, 具有重要的工程应用价值。

Single-cell Raman spectroscopy (SCRS) technology has the advantages of being rapid, sensitive, and label-free to study cell structure at the single-cell level. A cell growth detection method based on Spectral Clustering and SCRS was proposed in this paper. SCRS data of 600 synchronous culture fermentation-engineered bacteria E. Coli were collected as experimental data, and SCRS data of 300 fermentation-probiotic bacteria-Bacillus subtilis, were collected to verify the methods applicability. Firstly, the growth curve of OD600 was measured for the synchronously cultured colonies as growth period labels at the microbial population level. Secondly, t-SNE was applied to visualize the SCRS data of the population cells, guiding Spectral Clustering to cluster the high-dimensional SCRS data. Silhouette Coefficient and CH index were applied to evaluate the best clusters and assign labels to each SCRS data cluster. Finally, the intersection of SCRS data cluster labels and growth period labels was fitted by cubic spline interpolation to accurately identify the heterogeneous growth period data co-existing in the population and achieve accurate identification of growth periods of single-celled microorganisms. The results showed that the cell growth analysis method based on spectral clustering and SCRS could effectively detect 9% and 4.3% heterogeneous data of the optimal clusters in the three growth periods by using a 2-dimensional embedding space dimension and nearest neighbor-based spectral clustering similarity calculation method according to the cell growth curve of synchronous culture population. The study proposed a method of unsupervised detection of single-cell growth, with the help of spectral clustering without tags, can directly according to the features of SCRS data modeling, and can be of the arbitrary shape of high-dimensional SCRS data clustering and the advantages of fast convergence, realized with two kinds of fermentation engineering bacteria and probiotic fermentation cells lag, the accuracy of logarithmic phase and stable phase identification. In a real sense, it can detect cell growth from the single cell level and provide more accurate and real-time control guidance for fermentation engineering, which has important engineering application value.