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

• Published in: Genes to cells : devoted to molecular & cellular mechanisms Vol. 30; no. 5; p. e70050
• Main Authors: Ikura, Masae, Ikura, Tsuyoshi, Furuya, Kanji
• Format: Journal Article
• Language: English
• Published: England 01.09.2025
• Subjects: Cell Cycle Proteins - genetics; Cell Cycle Proteins - metabolism; Cell Line, Tumor; DNA Damage; Fluorescence; Histones - metabolism; Humans; Machine Learning; Neoplasms - metabolism; Radiation Tolerance
• ISSN: 1365-2443; 1365-2443
• DOI: 10.1111/gtc.70050

The thymidine analog EdU (5-ethynyl-2-deoxyuridine) is incorporated into DNA during replication and has traditionally been used as a marker of S-phase cells. In this study, we discovered that EdU fluorescence images display substantial cell-to-cell variability, which could be classified into multiple clusters by unsupervised machine learning. This suggests that seemingly random EdU patterns contain reproducible, computationally recognizable features. Building on our observation that distinct patterns emerged in response to radiation stress, we investigated whether radioresistant cancer cells exhibit specific EdU signatures. Analysis of PLK1-overexpressing cells, which acquire radioresistance through altered DNA replication, revealed radiation-induced EdU patterns distinct from control cells. Prompted by the observation that these cells displayed markedly enlarged and intensified γ-H2AX foci, a marker of DNA damage, we employed a supervised machine learning model based on γ-H2AX patterns to isolate the radioresistant cell subpopulation. We then extracted the EdU signals from these isolated cells and, through further unsupervised machine learning, successfully identified a characteristic pattern specific to radioresistance. This establishes a machine learning framework capable of extracting universal rules from the dynamic networks that vary among individual cells, which provides a novel platform for a screening system to identify molecules involved in radioresistance, focusing on cancer heterogeneity.