Detecting and classifying metabolic activity of by DO-probed single-cell Raman spectroscopy and machine learning.

The metabolic activity of pathogens poses a substantial risk across diverse domains, including food safety, vaccine development, clinical treatment, and national biosecurity. Conventional subculturing methods typically require several days and fail to detect metabolic activity promptly, limiting their application in many areas. Consequently, there is an urgent need for a method capable of rapidly and accurately detecting this activity. This study builds upon an investigation of the effects of DO on (), utilizing DO-probed single-cell Raman spectroscopy to detect the metabolic activity of by the Carbon-Deuterium ratio (C-D). Then, it evaluates the performance of various machine learning models in classifying the metabolic states of the pathogen. Medium DO concentration below 50 % has no significant impact on the growth and reproduction of or on the classification of metabolic states of based on the fingerprint region by machine learning models. Additionally, as the metabolic activity of decreases, both the C-D and the rate of viable cells also gradually decrease. The support vector machine model demonstrated an accuracy of 99.82 % in classifying viable and dead , while the linear discriminant analysis model demonstrated an accuracy of 99.92 % in classifying exhibiting distinct metabolic activities. Therefore, DO-probed single-cell Raman spectroscopy, combined with high-throughput technology, can rapidly, non-destructively, and accurately detect pathogen metabolic activity, offering valuable applications across multiple fields.