Single-cell Raman spectroscopy-machine learning combination: Rapid and accurate strain-level identification of...

Rapid and accurate strain-level identification of Bifidobacterium animalis is critical for probiotic quality control and intellectual property protection in the food industry. In this study, a novel identification method for Bifidobacterium animalis subsp. lactis J12 was established by integrating single-cell Raman spectroscopy (SCRS) with machine learning (ML) algorithms. A total of 500-1000 valid single-cell Raman spectra were collected for each of five Bifidobacterium animalis subsp. lactis strains and seven heterologous lactic acid bacterial strains. The SCRS data were processed and optimized using t-distributed stochastic neighbor embedding (t-SNE), linear discriminant analysis (LDA), and support vector machine (SVM) algorithms. Results showed that stationary-phase cells exhibited the most distinctive chemical fingerprints, enabling the SVM model to identify J12 among the five target strains with an accuracy of over 99%. The established method displayed high robustness: classification accuracy and recall for lyophilized bacterial powders both exceeded 99.8%; the predicted proportions of strains in artificial mixed samples were highly consistent with theoretical values, with a mean absolute error (MAE) of <0.5% and a correlation coefficient (R²) of >0.998; and the identification accuracy for J12 isolates from commercial yogurt remained above 99%. Comparative genomics analysis confirmed that the unique Raman phenotypic characteristics of J12 were correlated with its specific genomic features, verifying the method's reliability at the genetic level. This label-free and pure culture-free SCRS-ML approach provides a rapid and high-precision tool for strain-specific identification in the probiotic industry, and fills the technical gap in rapid detection of highly homologous B. animalis strains.