Exp Mol Med. 2026 Apr 21. doi: 10.1038/s12276-026-01674-7. Online ahead of print.
ABSTRACT
The rapid and accurate detection of multiple cancers presents considerable challenges, especially for stage I disease, due to the low concentration and heterogeneous nature of circulating tumor DNA. Here we introduce an enhanced multicancer screening assay that integrates whole-genome methylation sequencing with an innovative multimodal analytical framework for cell-free DNA. The ensemble machine learning model integrates four specific cell-free DNA characteristics: average methylation fraction, copy number variation, fragment size ratio and fragment size distribution. The model underwent testing on 1415 samples, encompassing eight primary cancer types and healthy controls. The sensitivity was 93.2%, and the specificity was 95%. The test demonstrated effectiveness in detecting cancers at early stages. The sensitivity was 92.3% for stage I and 92.2% for stage II. The multimodal technique successfully combined average methylation fraction's sensitivity to early epigenetic signals with fragmentomic characteristics. This facilitated the differentiation between healthy individuals and those with early stage cancer. The model achieved an accuracy rate of 85.7% in the top 2 category for correctly identifying the tissue of origin. The results confirm that whole-genome methylation sequencing-based multimodal analysis can improve multicancer early detection technology and revolutionize cancer screening methods.
PMID:42014847 | DOI:10.1038/s12276-026-01674-7