Pancreatic ductal adenocarcinoma (PDAC) has one of the poorest prognoses of any cancer, largely because it is rarely caught early. Most patients are diagnosed at advanced stages when surgery is no longer possible, leaving clinicians with limited treatment options and patients with a median survival of less than a year.
The biology of PDAC makes early detection especially difficult. The pancreas sits deep within the abdomen, tumors grow silently without producing obvious symptoms, and current biomarkers such as CA 19-9 lack the sensitivity and specificity needed for reliable screening in the general population.
This study set out to use bioinformatics approaches to identify a novel panel of molecular biomarkers that could improve the accuracy of PDAC diagnosis, particularly in the early stages where intervention is most effective.
Weighted Gene Co-Expression Network Analysis (WGCNA) is a systems biology method that groups genes into modules based on their co-expression patterns across many samples. Genes within the same module tend to share biological functions or regulatory relationships, and hub genes with the highest connectivity are often the most biologically significant.
The researchers applied WGCNA to publicly available gene expression datasets from the Gene Expression Omnibus (GEO), comparing pancreatic cancer tissue samples against normal pancreatic tissue. This allowed them to identify modules of genes whose expression was consistently altered in PDAC.
From these modules, candidate hub genes were selected based on their module membership scores and gene significance values. A shortlist of the most promising candidates was taken forward for further validation using independent datasets and laboratory experiments.
The analysis identified four genes as top hub candidates: TSPAN1 (Tetraspanin 1), TMPRSS4 (Transmembrane Serine Protease 4), SDR16C5 (Short-Chain Dehydrogenase/Reductase Family 16C Member 5), and CTSE (Cathepsin E). Each of these genes showed significantly elevated expression in PDAC compared to normal pancreatic tissue.
These four genes were validated using multiple independent GEO datasets, confirming that their upregulation was a consistent and reproducible feature of PDAC. The genes were also assessed at the protein level using immunohistochemistry on tissue microarrays, which confirmed their overexpression in tumor tissue relative to adjacent normal tissue.
Functional enrichment analysis revealed that the hub genes were involved in processes critical to cancer biology, including cell migration, invasion, and proliferation, suggesting that they are not merely passive bystanders but active contributors to PDAC pathogenesis.
To evaluate the diagnostic potential of the four hub genes, the researchers trained multiple machine learning classifiers using the gene expression levels of TSPAN1, TMPRSS4, SDR16C5, and CTSE as input features. Classifiers tested included Support Vector Machine (SVM), Random Forest, Naive Bayes, and Neural Network models.
Each model was evaluated using standard cross-validation procedures to prevent overfitting and to ensure the results would generalize to new samples. Performance was measured by the area under the receiver operating characteristic curve (AUC-ROC), sensitivity, and specificity.
The models were trained on one cohort and then tested on independent external validation datasets, providing a rigorous assessment of how well the gene panel could distinguish PDAC cases from healthy controls in genuinely unseen samples.
The four-gene panel demonstrated impressive diagnostic accuracy. AUC values ranged from 0.87 to 0.98 across the different machine learning models and validation datasets, indicating that the panel can reliably distinguish PDAC from normal tissue.
The SVM and Random Forest models achieved particularly high performance, with sensitivity and specificity both exceeding 90% in several cohorts. This level of accuracy surpasses what is currently achievable with CA 19-9 alone, which typically provides an AUC of around 0.82 for PDAC diagnosis.
The consistency of results across multiple classifiers and independent validation datasets strengthens confidence that the four-gene panel captures a genuine and robust molecular signal of PDAC, rather than a statistical artifact of the training data.
TSPAN1 belongs to the tetraspanin family, which regulates cell adhesion, migration, and signaling. Its overexpression in various cancers has been linked to enhanced tumor invasion and metastasis, and in PDAC it may contribute to the highly aggressive growth pattern characteristic of this disease.
TMPRSS4 is a cell-surface serine protease that promotes epithelial-to-mesenchymal transition (EMT), a process by which cancer cells acquire migratory and invasive properties. Elevated TMPRSS4 has been reported in several gastrointestinal cancers and may facilitate the early spread of PDAC beyond the pancreas.
CTSE, an aspartic protease, has previously been identified in PDAC-related gene signatures. Its role in degrading extracellular matrix proteins may facilitate local tissue invasion, while SDR16C5 is less well characterized but appears to participate in metabolic reprogramming that supports rapid tumor growth.
The four-gene panel could potentially be incorporated into several types of clinical diagnostic workflows. Measurement of the panel's expression from tissue biopsies obtained via endoscopic ultrasound-guided FNA could complement histological assessment, particularly in cases where the biopsy material is insufficient for definitive diagnosis.
Looking further ahead, it may be possible to detect altered expression or related protein levels in liquid biopsy materials such as plasma or serum, which would allow non-invasive monitoring. However, this would require additional research to establish whether the panel's signal is detectable in blood samples.
The panel could also be valuable for identifying high-risk individuals among those with precursor lesions such as intraductal papillary mucinous neoplasms (IPMN), helping clinicians decide which patients need more intensive surveillance or earlier surgical intervention.
This study demonstrates that a four-gene panel comprising TSPAN1, TMPRSS4, SDR16C5, and CTSE, identified through WGCNA and validated by machine learning and immunohistochemistry, constitutes a promising diagnostic signature for pancreatic cancer.
The combination of bioinformatics discovery with experimental validation across multiple independent cohorts provides a strong foundation for future clinical studies. The next logical step is prospective validation in well-defined clinical cohorts to confirm the panel's performance in real-world diagnostic settings.
If further validated, this molecular panel could meaningfully improve early detection rates for PDAC, shifting more diagnoses to stages where curative-intent surgery remains possible and where long-term survival outcomes are substantially better.