Raman spectroscopy combined with machine learning for rapid detection of food-borne pathogens at the single-cell level

Abstract

Rapid detection of food-borne pathogens in early food contamination is a permanent topic to ensure food safety and prevent public health problems. Raman spectroscopy, a label-free, highly sensitive and dependable technology has attracted more and more attention in the field of diagnosing food-borne pathogens in recent years. In the research, 15,890 single-cell Raman spectra of 23 common strains from 7 genera were obtained at the single cell level. Then, the nonlinear features of raw data were extracted by kernel principal component analysis, and the individual bacterial cell was evaluated and discriminated at the serotype level through the decision tree algorithm. The results demonstrated that the average correct rate of prediction on independent test set was 86.23 ± 0.92% when all strains were recognized by only one model, but there were high misjudgment rates for certain strains. Therefore, the four-level classification models were introduced, and the different hierarchies of the identification models achieved accuracies in the range of 87.1%–95.8%, which realized the efficient prediction of strains at the serotype level. In summary, Raman spectroscopy combined with machine learning based on fingerprint difference was a prospective strategy for the rapid diagnosis of pathogenic bacteria.