Bioimage analysis for multiplexed FUCCI acquisitions powered by deep learning.

Area of Science:

• Cell Biology
• Biotechnology
• Computational Biology

Background:

• The FUCCI (Fluorescence Ubiquitous Chromatin-tagging) sensor is crucial for visualizing cell cycle phases in developing organisms.
• Accurate cell cycle decoding is difficult in live cell imaging due to low signal-to-noise ratios.
• Existing methods struggle with segmentation and classification of FUCCI signals under suboptimal imaging conditions.

Purpose of the Study:

• To develop an advanced deep learning approach for precise cell cycle analysis using FUCCI signals.
• To improve the segmentation and classification of FUCCI nuclei in challenging imaging environments.
• To enable robust automated cell tracking and pseudotime analysis for cell cycle studies.

Main Methods:

• Development of deep learning networks integrating FUCCI signals with an alpha-tubulin reporter.
• Application of the networks for segmentation and classification of FUCCI nuclei.
• Implementation of dynamic time warping for cell cycle pseudotime analysis from incomplete tracks.
• Provision of pre-trained networks for multichannel FUCCI analysis.

Main Results:

• The deep learning approach significantly outperforms existing methods in segmenting and classifying FUCCI nuclei.
• High-accuracy segmentation facilitates robust automated tracking of cell cycle progression.
• The dynamic time warping analysis successfully determines cell cycle pseudotime and detects cell cycle arrest.
• The method demonstrates effectiveness even in low signal-to-noise conditions.

Conclusions:

• The developed deep learning tool provides a powerful and accurate method for cell cycle analysis using FUCCI sensors.
• This approach overcomes limitations of current techniques in low signal-to-noise live cell imaging.
• The tool is applicable to diverse research areas including cancer research, developmental biology, and mechanobiology.