Efficient Detection of Longitudinal Bacteria Fission Using Transfer Learning in Deep Neural Networks.

Area of Science:

• Microbiology
• Computational Biology
• Machine Learning

Background:

• Microscopic cell counting is crucial for measuring microbial growth and classifying bacteria.
• Automated methods exist, but manual classification persists for bacteria with longitudinal division, such as Candidatus Thiosymbion oneisti.
• Accurate identification of division types is vital for understanding bacterial cell cycles.

Purpose of the Study:

• To develop an automated method for classifying longitudinal bacterial cell division using machine learning.
• To reduce the manual labor associated with classifying bacteria cell division.
• To enable automatic labeling of bacteria division types from microscopic images.

Main Methods:

• Utilized a machine learning approach, specifically a residual network, for binary classification.
• Employed transfer learning to train the classification model efficiently.
• Developed a process to detect and segment individual bacteria from microscopic colony images.

Main Results:

• Successfully automated the classification of longitudinal bacterial division.
• Transfer learning enabled model training in fewer epochs compared to traditional training.
• The method has the potential to significantly decrease manual classification efforts.

Conclusions:

• The developed machine learning model effectively automates the classification of longitudinal bacterial division.
• This approach offers a faster and more efficient alternative to manual cell counting for specific bacteria.
• The technique is applicable for automatic labeling of bacteria division post-segmentation from microscopic data.