Fragmentomic liquid biopsy enables early breast cancer detection, molecular subtyping and lymph node assessment

Breast cancer remains a leading global health concern in women, while screening is still limited by imaging accessibility and reduced sensitivity in dense breasts. Here we conduct a multicenter case-control study including 503 breast cancer patients and 289 benign controls to develop TuFEst, a machine learning model based on genome-wide cell-free DNA fragmentomic features. TuFEst achieves high sensitivity (95%) and specificity (78.3%) for early cancer detection and reliably identifies malignancies missed by conventional imaging. Extension of this framework enables non-invasive molecular subtyping (TuFEst-MS) and lymph node status prediction (TuFEst-LN), with strong performance in independent validation cohorts and imaging-pathology discordant cases. Transcriptomic profiling of paired bulk tumor samples (n = 79) demonstrates that elevated TuFEst-derived cancer scores reflect tumor aggressiveness and immune-related biological programs. Together, these findings support cfDNA fragmentomics as an integrated liquid biopsy strategy for breast cancer management, enabling concurrent detection, molecular subtyping, and lymph node evaluation with potential clinical utility. By analysing genome-wide cfDNA fragmentation, a prospective case-control study develops a machine learning model that sensitively detects early breast cancer, classifies tumour subtypes, and assesses lymph node involvement from a single blood draw