Circulating cell-free mtDNA fragmentomics for early detection of gastric cancer and precancerous lesions.

Gastric cancer (GC) requires non-invasive early detection beyond insensitive serum biomarkers and invasive endoscopy. Circulating cell-free mitochondrial DNA (ccf-mtDNA) fragmentomics offers liquid biopsy potential. Using capture-based deep sequencing in 1,951 participants across multiple centers, we develop a random forest-based GC detection (GD) model. The GD model achieves high accuracy for GC (sensitivity >90%, specificity >93%). In a nested case-control study (42 GC cases, 42 matched controls), it detects preclinical GC with time-dependent sensitivity from 62.50% (>3 years pre-diagnosis) to 85.71% (<2 years) and stable specificity (88.89%-100%). For high-grade dysplasia, the model shows superior sensitivity to serum biomarkers (83.10% vs. 25.00%, p < 0.001). In other gastrointestinal cancers (n = 194), the area under the curve is 0.7477 with a specificity of 70.10%. ccf-mtDNA fragmentomics thus enables the accurate, non-invasive detection of GC and its precursors, supporting clinical diagnosis and risk management.