bims-fragic 2026-08-09 papers

Issue of 2026–08–09
three papers selected by
Laura Mannarino, Humanitas Research

Mol Biomed. 2026 Aug 05. pii: 125. [Epub ahead of print]7(1):

Integrative cfDNA profiling from low-pass whole-genome sequencing enables tissue-of-origin prediction in cancer.

Cancer type classification is challenging due to tumor heterogeneity and undefined tissue of origin (TOO), particularly in cancers of unknown primary (CUP) and multiple primary cancers (MPC). Accurate TOO identification is critical for guiding treatment and prognosis. We developed a stacked ensemble machine learning classifier that integrates 11 multidimensional cfDNA features spanning genomic, fragmentomic, methylation/repeat, and microbial signals. Base models were constructed using five algorithms, including Deep Learning, Distributed Random Forest, Gradient Boosting Machine, Generalized Linear Model, and XGBoost, within a five-fold cross-validation framework, and their predictions were aggregated into a final ensemble optimized for top-1 accuracy. The classifier achieved robust performance across 17 cancer types, with top-1 and top-2 accuracies of 78% and 89% in the training cohort (n = 1,814), and 80% and 90% in an independent validation cohort (n = 1,221). Notably, predictive performance was retained in samples with low tumor fraction (71% top-1, 85% top-2). Sensitivity varied across tumor types, with the highest performance observed in head and neck and colorectal cancers. Among CUP cases, 11 of 15 (73.3%) predictions matched clinically inferred primary sites based on multimodal diagnostics. Feature importance analysis identified nucleosome positioning, fragment size distribution, and repeat elements as key contributors to model performance. Collectively, this cfDNA-based classifier provides a robust and non-invasive approach for accurate cancer type identification and has the potential to support clinical decision-making.

Keywords: Multi-cancer classification; Tissue-of-origin; Whole genome sequencing; cfDNA profiling

DOI: https://doi.org/10.1186/s43556-026-00497-2

Mil Med. 2026 Aug 01. 191(Supplement_1): 673-682

Artificial Intelligence/Machine Learning-Driven cfDNA Methylation Blood Test for Early Cancer Detection in Military Personnel With Environmental Toxin Exposure.

Mythili Subharam, Ryan Koehler, Tejas Sreedhar.

INTRODUCTION: Military personnel face heightened cancer risks from exposure to burn-pit emissions, per- and polyfluoroalkyl substances (PFAS), jet fuel, and radiation. Veterans show elevated malignancy rates, and active-duty personnel remain vulnerable because of ongoing operational exposures, underscoring the need for early detection. Traditional screening methods are invasive and have limited sensitivity for early-stage disease. This study evaluates the feasibility of artificial intelligence/machine learning (AI/ML)-based cell-free DNA (cfDNA) methylation analysis for early cancer detection using publicly available civilian datasets, to inform future application in high-risk military populations.
MATERIALS AND METHODS: A Machine-Learning driven cfDNA methylation classifier framework was developed for early, minimally-invasive detection of cancers associated with toxic and occupational exposures. The platform uses bisulfite-sequenced data to identify differential methylation signatures that distinguish early stage hepatocellular carcinoma (HCC) from chronic liver disease and healthy controls. Proof-of-concept studies were conducted on public datasets for hepatocellular carcinoma (HCC) and esophageal cancer. Three cfDNA datasets (GSE93203, GSE63775, and PRJCA001372) were used for model training and k-fold cross-validation, and an independent WGBS liver-tissue dataset (PRJNA984754) served as a blind cross-assay validation set. Framework extensibility to other cancers was assessed through the EpiPanGI-Dx esophageal-cancer dataset (ESCC training, EAC validation).
RESULTS: Across 3 HCC models, internal cross-validation achieved 84%-94% accuracy (AUC 0.80-0.88). On the independent WGBS dataset validated against the 3 trained models, classifiers generalized with 83%-100% accuracy (AUC 0.80-1.00). Clinical validation using biobank cfDNA plasma samples from early-stage hepatocellular carcinoma and cirrhosis controls is ongoing. The esophageal cancer model reproduced published performance (AUC 0.94 for ESCC, 0.90 for EAC), demonstrating generalizability beyond HCC.
CONCLUSION: The cfDNA methylation-based AI/ML platform is promising for early multi-cancer detection in environmentally exposed military populations. Early results demonstrate feasibility and justify expansion to additional cancers, and larger clinical validation studies, potentially enhancing survivability through earlier intervention.

DOI: https://doi.org/10.1093/milmed/usag226

Front Cell Dev Biol. 2026 ;14 1874565

Liquid biopsy biomarkers for cancer detection, treatment monitoring, and clinical outcome prediction.

Akshee Batra, Xena Zheng, Dan Morgenstern-Kaplan, Carey C Thomson, Gilberto Lopes, Chinmay Jani.

Liquid biopsy now provides minimally invasive access to tumor-derived genomic and epigenetic information across the lung cancer continuum, and its clinical role continues to expand. This review examines that role across cancer detection (screening and diagnosis), treatment monitoring (advanced-disease genotyping, minimal residual disease (MRD) assessment, and resistance profiling at progression), and clinical outcome prediction. Plasma-based genotyping is now well established in advanced non-small cell lung cancer (NSCLC), while circulating tumor DNA (ctDNA)-based MRD detection in the curative-intent setting has accumulated a substantial evidence base over the past 5 years. Cell-free DNA (cfDNA) methylation, fragmentomics, and circulating tumor RNA (ctRNA) are emerging as complementary modalities, particularly when tumor shedding is low. We also consider concordance between liquid and tissue biopsies, the use of cerebrospinal fluid (CSF) ctDNA in central nervous system (CNS)-involved disease, and the practical issues of cost, reimbursement, and access that shape clinical adoption. The current state of the field can be framed across three tiers of evidence, with established applications, applications under prospective evaluation, and applications not yet ready for routine clinical use. No multi-cancer early detection (MCED) test has shown a mortality benefit to date, and ctDNA-guided treatment changes in metastatic disease still lack randomized overall-survival data.

Keywords: circulating tumor RNA; circulating tumor cell (CTC); liquid biopsy; lung cancer; methylation and prognosis

DOI: https://doi.org/10.3389/fcell.2026.1874565