Rapid classification of bacteria by a portable Raman spectrometer and machine learning

Efficient, accurate, and early identification of plant pathogens is crucial for reducing disease spread and ensuring food security. The development of rapid diagnostic methods based on Raman spectroscopy (RS) coupled with machine learning (ML) holds great potential to enable prompt and targeted responses. To enhance the practical applicability of RS for the identification of plant pathogenic bacteria, we investigated the use of a hand-held RS instrument coupled with ML techniques to differentiate isolates belonging to the Pseudomonas spp., Xanthomonas spp., and Erwinia spp. genera. 332 Raman spectra were acquired directly on bacterial colonies grown on a solid medium. After comparing different algorithms' performances, PLS-DA models were built, enabling the identification of various isolates at the genus, species, and pathovars levels, with accuracies ranging from 83% up to 99%. This easy and fast approach provides the opportunity to develop a widely accessible database of bacterial spectra, thereby facilitating non-destructive and rapid microorganisms' classification, ultimately aiding in the prevention and management of crop diseases.