A Machine Learning Framework Using Urinary Biomarkers for Pancreatic Ductal Adenocarcinoma Prediction With Post Hoc Validation via Single-Cell Transcriptomics

This study presents a machine learning framework for predicting pancreatic ductal adenocarcinoma (PDAC) using urinary biomarkers, with post hoc validation through single-cell transcriptomics. Available urinary biomarkers were incorporated into a demographic-informed preprocessing approach, followed by predictive model development using normalization and feature-learning techniques. The framework was further validated using single-cell RNA sequencing (scRNA-seq) data to assess gene-expression patterns of identified biomarkers within pancreatic tumor samples. The analysis demonstrated the significance of urinary biomarkers for PDAC prediction and highlighted their biological relevance through transcriptomic validation. Multiple classification models and parameter combinations were evaluated, with deep learning approaches outperforming traditional machine-learning methods. The predictive model achieved high accuracy, and the framework provided insights into demographic influences without compromising predictive performance. The study supports the potential application of urinary biomarkers as a non-invasive tool for early PDAC detection and emphasizes the value of integrating transcriptomic validation into biomarker-based predictive modeling.