AI detects a distributed blood metabolomic Systemotype associated with early stage ovarian cancer

Early detection of ovarian cancer remains a clinical challenge because available blood biomarkers lack the sensitivity and specificity required for population screening. We tested whether early-stage ovarian cancer is associated with a distributed physiological state in the circulating metabolome. We analyzed untargeted metabolomic profiles from two independent retrospective cohorts: 91 serum samples (59 ovarian cancer, 32 healthy controls) and 83 plasma samples (63 ovarian cancer, 20 healthy controls). Assay-specific boosted decision trees were trained and evaluated independently within each cohort using five-fold cross-validation repeated over 50 randomized rounds. At selected operating points, mean cross-validated sensitivity and specificity were 99.0% and 99.8% in serum and 97.7% and 99.7% in plasma. Restricting inputs to strongly dysregulated features did not improve the overall sensitivity-false-positive-rate trade-off, and smaller panels reduced sensitivity. The cohorts shared 239 concordantly altered annotated features spanning lipid, amino-acid, steroid, central-carbon, and redox metabolism. These findings are consistent with a distributed metabolic response involving tumor and host, although tissue contributions were not measured. We propose that the classifier recognizes a metabolomic Systemotype, an integrated physiological state reflected in circulating metabolites. The results support further investigation of this framework.