A Noninvasive and Screening miRNA Signature for Gastrointestinal Cancer

This study will establish a comprehensive, retrospective, international multi-center cohort consisting of peripheral blood samples from participants with major gastrointestinal cancers-including hepatocellular carcinoma (HCC), cholangiocarcinoma (CCA), pancreatic ductal adenocarcinoma (PDAC), esophageal squamous cell carcinoma (ESCC), gastric cancer (GC), and colorectal cancer (CRC)-as well as non-malignant controls. Small RNA sequencing will be performed to generate high-resolution circulating miRNA expression profiles. During the discovery phase, the investigators will conduct rigorous preprocessing, normalization, batch effect correction, and differential expression analyses to identify circulating miRNAs associated with malignant transformation across GI cancer types. Machine learning-based feature selection (e.g., LASSO, mRMR, ensemble methods) and classifier development (e.g., SVM, Random Forest, XGBoost) will then be used to derive a minimal yet robust miRNA panel capable of optimally distinguishing cancer from non-cancer. During the modeling and evaluation phase, the identified miRNA signature will undergo multi-center training and validation across international cohorts to ensure robustness across geographic regions, sequencing platforms, and clinical demographics. Beyond binary classification, the investigators will assess the panel's ability to discriminate among specific GI cancer subtypes, thereby supporting differential diagnosis and tumor-origin inference. Model performance will be evaluated using AUROC, sensitivity at clinically meaningful specificity thresholds, early-stage detection capability, and calibration in independent validation cohorts. Through this sequential discovery → modeling → multi-center validation framework, the investigators aim to develop a noninvasive circulating miRNA panel that (1) accurately distinguishes cancer from non-cancer individuals and (2) differentiates among multiple gastrointestinal cancer types, thereby providing a clinically scalable solution for early cancer detection and population-level screening.