Machine Learning Extracts Radiation Resistant-Specific EdU Fluorescence Pattern in Cancer Cells.

Area of Science:

• Cell Biology
• Genomics
• Bioinformatics

Background:

• 5-ethynyl-2-deoxyuridine (EdU) is a thymidine analog incorporated into DNA during replication, marking S-phase cells.
• EdU fluorescence images show significant cell-to-cell variability, with patterns classifiable by machine learning.
• Distinct EdU patterns emerge under radiation stress, suggesting potential for identifying radioresistant cells.

Purpose of the Study:

• To investigate if radioresistant cancer cells exhibit specific EdU signatures.
• To develop a machine learning framework for identifying radioresistance-associated EdU patterns.
• To establish a novel screening platform for molecules involved in radioresistance.

Main Methods:

• Unsupervised and supervised machine learning algorithms were applied to analyze EdU fluorescence images.
• PLK1-overexpressing cells, known for radioresistance, were analyzed for distinct EdU patterns.
• γ-H2AX foci, a marker of DNA damage, were used to isolate radioresistant subpopulations for EdU signal extraction.

Main Results:

• Radiation stress induced distinct EdU patterns in PLK1-overexpressing radioresistant cells compared to controls.
• A supervised machine learning model utilizing γ-H2AX patterns successfully isolated radioresistant cells.
• Unsupervised machine learning identified a characteristic EdU pattern specific to radioresistance within the isolated cell subpopulation.

Conclusions:

• Machine learning can extract reproducible features from variable EdU patterns, revealing insights into cellular responses.
• A novel machine learning framework was established to identify specific EdU signatures associated with radioresistance.
• This approach provides a new platform for screening molecules that modulate radioresistance in cancer, addressing cancer heterogeneity.