A deep learning approach for time-consistent cell cycle phase prediction from microscopy data.

Area of Science:

• Cell Biology
• Molecular Biology
• Bioimaging

Background:

• The cell cycle comprises regulated stages (G1, S, G2, M) crucial for cell growth, DNA replication, and division.
• Identifying cell cycle phases typically requires specific markers, which can interfere with other experimental reporters in imaging assays.

Purpose of the Study:

• To develop a method for inferring cell cycle phase from commonly used DNA fluorescent reporters, bypassing the need for dedicated cell-cycle markers.
• To enable cell cycle analysis in high-content screening datasets not originally designed for this purpose.

Main Methods:

• A Variational Auto-Encoder (VAE) model, termed CC-VAE, was developed.
• The VAE was enhanced with auxiliary tasks: predicting phase-specific marker intensity and enforcing temporal consistency via latent space regularization.
• The model was trained and validated on a large dataset of labeled HeLa Kyoto nuclear images.

Main Results:

• CC-VAE accurately classifies cell cycle phases using only standard DNA markers like SiR-DNA.
• The method effectively bypasses the need for additional, potentially interfering, phase-specific fluorescent markers.
• The model demonstrates high accuracy and applicability to diverse high-content screening datasets.

Conclusions:

• CC-VAE offers a practical and efficient solution for cell cycle phase determination in biological imaging.
• This method expands the utility of existing experimental setups for cell cycle analysis without requiring specialized reagents.
• The developed model and associated dataset facilitate advancements in cell cycle research and high-content screening.