Rapid detection of microbiota cell type diversity using machine-learned classification of flow cytometry data

Area of Science:

• Microbiology
• Bioinformatics
• Machine Learning

Background:

• Complex microbial communities are typically studied using high-throughput sequencing and bioinformatics.
• Current methods can be time-consuming and may not capture the full diversity of microbial populations.

Purpose of the Study:

• To develop and validate a machine learning algorithm, CellCognize, for rapid cell type diversity quantification from multidimensional flow cytometry data.
• To demonstrate CellCognize's ability to analyze microbial community shifts and productivity.

Main Methods:

• Trained neural networks with 32 microbial cell and bead standards for supervised cell type recognition.
• Validated classifiers in silico on known microbiota, achieving an average of 80% prediction accuracy.
• Applied CellCognize to detect microbial community shifts upon chemical amendment and quantify population growth.

Main Results:

• CellCognize achieved 80% prediction accuracy in silico validation.
• The algorithm detected microbial community shifts comparable to 16S-rRNA-amplicon sequencing.
• CellCognize provided biomass productivity estimates similar to 14C-substrate incorporation methods.

Conclusions:

• CellCognize offers a rapid and complementary approach to sequencing-based methods for routine microbial cell diversity analysis.
• The pipeline is adaptable for optimizing cell recognition in recurring microbiota types, including those in human health and engineered systems.