Machine learning and gene network integration reveal prognostic subnetworks and biomarkers in pancreatic cancer.

Background: Pancreatic cancer has a high mortality rate and lacks early detection markers. Advanced methods, such as machine learning (ML) and network analysis, identify central cancer networks with potential diagnostic and prognostic biomarkers, leading to improved tumor targeting strategies.

Methods: We systematically collected pancreatic cancer transcriptome datasets from the databases TCGA, GTEx, and GEO. Weighted gene co-expression network analysis (WGCNA) identified gene modules associated with clinical traits. Multiple machine learning-based feature selection methods (Random Forest, Support Vector Machine, LASSO, ReliefF) and differential gene expression (DGE) analysis prioritized candidate genes. Functional enrichment (Gene Ontology and KEGG pathway database) examined biological processes involved in tumor progression and immune evasion. Survival analyses evaluated prognostic significance.

Results: WGCNA identified pancreatic cancer networks from key gene modules strongly associated with cancer stage and survival. Common biomarkers, including transcripts from genes ANLN, GPRC5A, KLF6, MUC1, and PHF20, demonstrated significant diagnostic and prognostic potential as shown by ML, WGCNA, DGE, and survival analyzes. In vitro validation was performed for proteins mucin1 and CD44 in patient samples and tissue models.

Conclusion: This study identified novel regulatory cancer networks and associated biomarkers for pancreatic cancer prognosis and diagnosis by integrating WGCNA with ML, DGE, pathway, and survival analyses. An interactive web portal to explore the full results and visualizations is available at pc-biomarkers.de. Future work will further validate these biomarkers to improve early detection, prognosis, and treatment strategies.