Pancreatic Cancer Detection and Differentiation from Chronic Pancreatitis: Potential Biomarkers Identified Through a High-throughput Multiplex Proteomic Assay and Machine Learning-based Analysis
- Author:
Young-gon KIM
1
;
Sang-Mi KIM
;
Soo-Youn LEE
Author Information
- Publication Type:Original Article
- From:Annals of Laboratory Medicine 2025;45(4):399-409
- CountryRepublic of Korea
- Language:English
-
Abstract:
Background:Pancreatic cancer (PC)-screening methods have limited accuracy despite their high clinical demand. Differential diagnosis of chronic pancreatitis (CP) poses another challenge for PC diagnosis. Therefore, we aimed to identify blood protein biomarkers for PC diagnosis and differential diagnosis of CP using high-throughput multiplex proteomic analysis.
Methods:Two independent cohorts (N = 88 and 80) were included, and residual serum samples were collected from all individuals (N = 168). Each cohort consisted of four groups: healthy (H) individuals and those with CP, stage I/II PC (PC1), or stage III/IV PC (PC2). Protein expression in the first cohort was quantified using the Olink Immuno-Oncology and Oncology 3 proximity extension assay (PEA) panels and was analyzed using machine-learning (ML)-based analyses. Samples in the second cohort were utilized to verify candidate biomarkers in immunoassays.
Results:Both the PEA and immunoassay results confirmed that previously recognized biomarkers, such as the mucin-16 and interleukin-6 proteins, were more highly expressed in the PC (PC1 and PC2) groups than in the non-PC (CP and H) groups. Several novel biomarkers for PC diagnosis were identified via ML-based feature extraction, including C1QA and CDHR2, whereas pro-neuropeptide Y (NPY) appeared to be a promising biomarker for the differential diagnosis of CP. Applying XGBoost classification incorporating the selected features resulted in an area under the curve of 0.92 (0.85–0.98) for differentiating the PC group from the CP and H groups.
Conclusions:Promising blood biomarkers for PC diagnosis and differential diagnosis of CP were identified using a PEA platform and ML techniques.
