Multiomic serum profiling identifies candidate signatures for classification and stage stratification of pancreatic ductal adenocarcinoma
The Journal of Precision Medicine: Health and Disease, 2026
Ganguly S., Devasahayam Arokia Balaya R., Byeon S., Kandasamy R.
| Disease area | Application area | Sample type | Products |
|---|---|---|---|
Oncology | Patient Stratification | Plasma | Olink Explore 3072/384 |
Abstract
Background
Pancreatic ductal adenocarcinoma (PDAC) is a lethal health issue with a dismal prognosis, primarily due to the high proportion of patients diagnosed at advanced stages. While serum biomarkers for PDAC have been discovered to differentiate PDAC patients from controls, a classifier to differentiate among PDAC stages has not yet been reported. Stage-specific serum biomarker panels could improve clinical stratification and personalized medicine.
Methods
Using a recently published serum multiomics dataset comprising 1463 proteins, 863 lipids, and 1064 metabolites from Stage III PDAC (n = 29), Stage IV PDAC (n = 29), and control subjects (n = 30), we built a hierarchical classifier and evaluated its performance over 10 repeats of stratified 5-fold cross-validation. Level 1 separated PDAC from Controls using L2-penalized logistic regression, and Level 2 separated Stage III from Stage IV among subjects classified as PDAC using ridge regression on the lipidome composition.
Results
The hierarchical classifier achieved mean subject-level AUCs of 0.973 in the Controls vs. PDAC level, and 0.868 in the Stage III vs. Stage IV PDAC level, with a mean end-to-end 3-class accuracy of 79.1%. 16 biomarkers selected for Level 1 in every training set were associated with biological processes including metabolic reprogramming, immune evasion, and release of tumor and stromal proteins. Interpreting the Level 2 classifier showed lipidome remodeling between Stage III and Stage IV PDAC.
Conclusion
Serum multiomics profiles contain stage-associated signatures that require validation in independent cohorts. This framework provides a foundation for future biomarker-based classifiers for early detection, staging, and prognostic prediction in PDAC.