Rapid Classification of Pantoea spp. via Raman Flow Cytometry

Microorganisms play pivotal roles in ecosystems, where precise taxonomic identification is fundamental to understanding and harnessing their functions. Conventional microbial classification, however, relies on pure-culture techniques that suffer from prolonged cultivation cycles, low throughput, high costs, and limited resolution. To address these constraints, we developed an integrated platform combining positive dielectrophoresis-activated Raman-activated cell sorting (pDEP-RACS) with deep ResNet. Using the ecologically versatile and taxonomically challenging genus Pantoea as a model, we constructed a reference Ramanome database by pDEP-RACS that consists of 180 000 single-cell Raman spectra (SCRS) from 12 Pantoea species (22 strains) and two phylogenetically related species and established a classification model by ResNet-18. The model achieved optimal performance 96.9% mean accuracy and 97.3% recall for 24 SCRS colony isolates. Cross-batch validation shows the highest reproducibility in nutrient-starved samples, with 87.9% accuracy postpreprocessing with reduced batch effects. For optimal accuracy (97.6% ± 2.0%), classification accuracy plateaus at >1,500 SCRS of Raman detection depth. In synthetic communities, the model shows ≤3.21% absolute abundance error for species identification. For the rice seed microbiome, a good consistency was observed between Raman-derived Pantoea abundance (34.8%) and 16S rRNA sequencing results (45%). This platform enables species- and strain-level classification of Pantoea spp. cultures, acquiring >7,200 SCRS per hour to facilitate rapid identification in both synthetic microbial communities and field-derived samples.