Hua Bao

Preoperative discrimination of ovarian cancer from benign masses poses significant challenges, particularly in cases where serum biomarkers yield negative or nonspecific results. This multicenter observational diagnostic modeling study analyzed plasma cell-free DNA (cfDNA) from 743 participants across training, internal test, external test, and two exploratory cohorts using low-pass whole-genome sequencing. Four fragmentomic feature classes—fragment size patterns, nucleosome footprint-related features, copy-number variations, and repeat element patterns—were incorporated into a stacked machine-learning model. Model development and cutoff selection were limited to the training cohort, followed by evaluation in the internal-test and external-test cohorts. The cfDNA model achieved areas under the receiver operating characteristic curve (AUCs) of 0.930 (95% confidence interval [CI], 0.894–0.966) and 0.932 (95% CI, 0.891–0.972) in the internal-test and external-test cohorts, respectively. Incorporation of cfDNA with cancer antigen 125 (CA125) and human epididymis protein 4 (HE4) increased AUCs to 0.964 (95% CI, 0.938–0.990) and 0.964 (95% CI, 0.938–0.991). At cutoffs derived from the training cohort, the combined model maintained sensitivities of 88.4% and 89.2% while enhancing specificity from 77.9% to 88.4% and from 78.9% to 91.5%, respectively. The cfDNA model also effectively detected cancers with CA125 and/or HE4 levels below conventional thresholds, and post hoc analyses demonstrated complementary discrimination when cfDNA was added to the Risk of Ovarian Malignancy Algorithm. These findings advocate for prospective evaluation of integrated cfDNA fragmentomics as an adjunct for preoperative risk stratification of ovarian masses.

Ultra-low-pass whole-genome sequencing (ULP-WGS) of cell-free DNA (cfDNA) offers a cost-efficient strategy for cancer detection, but its clinical application is limited by extreme data sparsity and poor model generalization. We developed Fragmentia-AI™ WGS, a mutation-calling-independent framework that uses a transformer-based multiple-instance learning architecture with sequential fine-tuning across tumor fraction (TF) strata to extract latent cancer-associated signals from ULP-WGS data. Model performance was evaluated in multiple independent cohorts, including a pan-cancer test set covering 17 cancer types, an external public dataset generated on a different sequencing platform, and a technical variability cohort with heterogeneous pre-analytical and experimental conditions. Clinical relevance was assessed by correlating model predictions with progression-free survival (PFS) in patients with advanced non-small cell lung cancer receiving chemoimmunotherapy. Sequential fine-tuning across TF strata significantly improved performance in low-TF samples, achieving a 35.6% relative increase in AUC compared with high-TF-only training (0.884 vs. 0.652). In the independent test cohort, the model achieved an overall AUC of 0.930, with consistent performance across TF strata and cancer types. External validation confirmed robust cross-platform generalizability (AUC: 0.929; sensitivity: 0.78; specificity: 0.92). The model maintained stable classification performance despite score fluctuations associated with pre-analytical and technical variables. Importantly, model-negative status, defined as a prediction score below the training-derived cutoff, remained significantly associated with improved PFS compared with model-positive status (HR = 0.49, 95% CI: 0.29–0.82) after multivariable adjustment. Collectively, this framework enables robust cancer detection and clinically meaningful risk stratification from highly sparse cfDNA sequencing data.

This file contains supplementary figures illustrating the development and validation of the cfDNA-based pancreatic cancer detection model, including prediction score distributions across clinical subgroups, feature importance analyses, assessments of the limit of detection, and estimated screening benefits. It also includes representative medical imaging and pathological results of the pancreatic cancer related-cases identified in the prospective cohort.

This file includes supplementary tables summarizing cohort demographics, in silico and empirical estimates of the limit of detection (LOD), model performance across clinical and demographic subgroups, results from the non-pancreatic cancer disease cohort and prospective screening cohort, and simulated clinical impact in hypothetical populations.