Quantification of Myxococcus xanthus Aggregation and Rippling Behaviors: Deep-Learning Transformation of...

Area of Science:

• Microbiology
• Biophysics
• Computational Biology

Background:

• Myxococcus xanthus exhibits complex collective behaviors like fruiting body formation and ripple formation.
• Phase-contrast and fluorescence microscopy are crucial for observing these patterns but have limitations.
• Phase-contrast offers high contrast but loses cell density correlation; fluorescence provides density correlation but requires cell engineering and can be phototoxic.

Purpose of the Study:

• To develop a computational method combining the advantages of phase-contrast and fluorescence microscopy.
• To generate synthetic fluorescent images from phase-contrast images of Myxococcus xanthus.
• To enable accurate analysis of bacterial collective behaviors and pattern formation.

Main Methods:

• A generative adversarial network (GAN) based on the pix2pixHD algorithm was developed.
• The GAN was modified to include a histogram-equalized output for improved image generation.
• The model was trained on phase-contrast images and validated using both aggregate and ripple patterns.

Main Results:

• The GAN successfully converted phase-contrast images into synthetic fluorescent images.
• The generated images accurately represented aggregate positions and sizes, with minor boundary shifts.
• The method enabled accurate estimation of ripple pattern wavelengths after further training.

Conclusions:

• The developed GAN provides a powerful tool for analyzing bacterial collective behaviors and pattern formation.
• This approach overcomes limitations of traditional microscopy, offering a non-invasive method for high-resolution imaging.
• The technique is broadly applicable to studies of phenotypic behaviors and pattern formation in other biological systems.