A research team led by the Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, together with collaborators including eCyte and the Qingdao Institute of Bioenergy and Bioprocess Technology, Chinese Academy of Sciences, has developed a rapid, culture-light strategy for identifying Pantoea bacteria at both the species and strain levels. The study was published in Analytical Chemistry.
Precise identification of closely related microbes remains a longstanding challenge in environmental and agricultural microbiology. Conventional methods often depend on pure culture, require lengthy workflows, and may still deliver only limited taxonomic resolution. In the new study, the researchers combined Raman flow cytometry with a deep residual neural network to overcome these bottlenecks in the taxonomically difficult genus Pantoea.
Using FlowRACS, eCyte’s high-throughput Raman flow cytometry platform, the team built a reference Ramanome database containing 180,000 single-cell Raman spectra collected from 12 Pantoea species, 22 strains, and two closely related outgroup species. Based on these data, they trained a ResNet-18 classification model that achieved a mean accuracy of 96.9% and a recall of 97.3% for colony isolates. The platform also supported rapid acquisition at more than 7,200 single-cell Raman spectra per hour, enabling taxonomic analysis on a much shorter timescale than sequencing-based approaches.
The researchers further showed that the method remained robust across experimental batches after optimization of pretreatment and sampling depth. Classification performance plateaued at greater than 1,500 spectra, where accuracy reached 97.6% ± 2.0%. In synthetic microbial communities, the model reconstructed species abundance with an absolute error of 3.21% or less. In rice seed microbiome samples, Raman-based analysis measured Pantoea abundance at 34.8%, compared with 45% by 16S rRNA sequencing, indicating good consistency while also suggesting the Raman workflow may preferentially reflect metabolically active cells.
The authors say the study provides a new route toward rapid in situ identification of environmentally and agriculturally important microbes, without relying exclusively on conventional cultivation workflows. Beyond resolving a difficult bacterial genus, the work establishes a methodological foundation for phenotype-informed analysis of complex microbial communities. For eCyte, the study also highlights how FlowRACS can support not only functional cell screening, but also high-throughput single-cell microbial identification when coupled with machine learning.