Machine Learning Models for Segmentation and Classification of Cyanobacterial Cells.

Area of Science:

• Microbiology
• Computational Biology
• Biotechnology

Background:

• Timelapse microscopy enables single-cell study of cyanobacteria metabolism and physiology.
• Accurate identification of individual cyanobacteria cells in dense colonies is challenging for traditional segmentation methods due to low contrast.

Purpose of the Study:

• To develop and validate machine learning (ML) models for segmenting individual cyanobacteria cells and classifying cellular phenotypes.
• To introduce Cypose, a software package integrating ML models for enhanced cyanobacteria image analysis.

Main Methods:

• Developed ML-based segmentation models using the Cellpose framework.
• Implemented a convolutional neural network (Cyclass) for cellular phenotype classification.
• Compared ML models against traditional methods for accuracy and robustness.

Main Results:

• Cypose models demonstrated superior performance in segmenting individual cyanobacteria cells, including those with varied morphologies and differentiating live from lysed cells.
• The models proved robust against imaging artifacts like dust and cell debris.
• The classification model accurately identified different cellular phenotypes directly from images.

Conclusions:

• Cypose provides the first ML-based solution for accurate cyanobacteria cell segmentation and classification.
• These models significantly improve cell segmentation accuracy, enabling high-throughput analysis of dense and filamentous cyanobacteria populations.