Scholarlink

Background <p>Circulating cell-free DNA (cfDNA) is a promising biomarker for non-invasive early cancer detection. We integrated methylation entropy, a measure of DNA methylation heterogeneity, with other cfDNA features into a multimodal model to improve multi-cancer detection.</p> Methods <p>A customized panel named HYGEIA, containing abundant informative methylation CpG sites, was first developed. The panel’s performance was validated, and a method for calculating fragment-level methylation entropy was established. Subsequently, this panel was utilized to perform deep panel-targeted bisulfite sequencing on 521 plasma samples from 288 cancer patients and 233 non-cancer individuals. Participants were randomly allocated into training (197 cancer, 157 non-cancer) and testing (91 cancer, 76 non-cancer) sets for constructing the Enhanced Multi-Cancer Early Detection (EMCED) model. Upon validation of the EMCED model, a Tissue-of-Origin (TOO) model was further developed and evaluated across seven cancer types.</p> Results <p>A classifier based solely on methylation entropy achieved an AUC of 0.938 (sensitivity 78.0%, specificity 94.7%) in the testing set for multi-cancer detection. Building on this, two additional characteristics, DNA methylation levels and fragmentation features, were integrated into the model, resulting in the EMCED model, which achieved an AUC of 0.979 (sensitivity 93.4%, specificity 93.4%) in the testing set. Analysis of the EMCED model’s sensitivity by cancer type showed detection rates of 95.5% for colorectal cancer, 98.0% for lung cancer, 93.5% for gastric cancer, 83.8% for ovarian cancer, and 100% for liver, esophageal, and thyroid cancers. The TOO model, which incorporated all three feature types, methylation entropy, methylation levels, and fragmentation features, achieved an accuracy of 92.3% for classifying the tissue of origin among cancer cases in the testing set.</p> Conclusions <p>Methylation entropy is a robust cfDNA biomarker for distinguishing cancer from non-cancer. The multimodal model enables enhanced multi-cancer detection and accurate classification, supporting the integration of methylation entropy for improved non-invasive early cancer detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Methylation entropy as a novel dimension in liquid biopsy: enhanced multimodal framework for cancer detection and tissue-of-origin classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月19日

查看全文
  • Xiaoqiong Jia,
  • Yongjun Li,
  • Baochen Du,
  • Jie Zhao,
  • Xiuxiu Wang,
  • Hong Yang,
  • Qiang Zhao,
  • Qin Si,
  • Yuanyuan Guo,
  • Bateer Han,
  • Zhenghang Wang,
  • Junqing Liang,
  • Yongfei Peng,
  • Shuanping Liu,
  • Guangpeng Zhou,
  • Zhongwu Li,
  • Ziyu Li,
  • Xiangdong Fang,
  • Xiaoliang Han

摘要

Background

Circulating cell-free DNA (cfDNA) is a promising biomarker for non-invasive early cancer detection. We integrated methylation entropy, a measure of DNA methylation heterogeneity, with other cfDNA features into a multimodal model to improve multi-cancer detection.

Methods

A customized panel named HYGEIA, containing abundant informative methylation CpG sites, was first developed. The panel’s performance was validated, and a method for calculating fragment-level methylation entropy was established. Subsequently, this panel was utilized to perform deep panel-targeted bisulfite sequencing on 521 plasma samples from 288 cancer patients and 233 non-cancer individuals. Participants were randomly allocated into training (197 cancer, 157 non-cancer) and testing (91 cancer, 76 non-cancer) sets for constructing the Enhanced Multi-Cancer Early Detection (EMCED) model. Upon validation of the EMCED model, a Tissue-of-Origin (TOO) model was further developed and evaluated across seven cancer types.

Results

A classifier based solely on methylation entropy achieved an AUC of 0.938 (sensitivity 78.0%, specificity 94.7%) in the testing set for multi-cancer detection. Building on this, two additional characteristics, DNA methylation levels and fragmentation features, were integrated into the model, resulting in the EMCED model, which achieved an AUC of 0.979 (sensitivity 93.4%, specificity 93.4%) in the testing set. Analysis of the EMCED model’s sensitivity by cancer type showed detection rates of 95.5% for colorectal cancer, 98.0% for lung cancer, 93.5% for gastric cancer, 83.8% for ovarian cancer, and 100% for liver, esophageal, and thyroid cancers. The TOO model, which incorporated all three feature types, methylation entropy, methylation levels, and fragmentation features, achieved an accuracy of 92.3% for classifying the tissue of origin among cancer cases in the testing set.

Conclusions

Methylation entropy is a robust cfDNA biomarker for distinguishing cancer from non-cancer. The multimodal model enables enhanced multi-cancer detection and accurate classification, supporting the integration of methylation entropy for improved non-invasive early cancer detection.