Integrating traditional machine learning with qPCR validation to identify solid drug targets in pancreatic cancer: a 5-gene signature study

Abstract

Background: Pancreatic cancer remains one of the deadliest malignancies, largely due to its late diagnosis and lack of effective therapeutic targets. Materials and methods: Using traditional machine learning methods, including random-effects meta-analysis and forward-search optimization, we developed a robust signature validated across 14 publicly available datasets, achieving a summary AUC of 0.99 in training datasets and 0.89 in external validation datasets. To further validate its clinical relevance, we analyzed 55 peripheral blood samples from pancreatic cancer patients and healthy controls using qPCR. Results: This study identifies and validates a novel five-gene transcriptomic signature (LAMC2, TSPAN1, MYO1E, MYOF, and SULF1) as both diagnostic biomarkers and potential drug targets for pancreatic cancer. The differential expression of these genes was confirmed, demonstrating their utility in distinguishing cancer from normal conditions with an AUC of 0.83. These findings establish the five-gene signature as a promising tool for both early, non-invasive diagnostics and the identification of actionable drug targets. Conclusion: A five-gene signature is established robustly and has utility in diagnostics and therapeutic targeting. These findings lay a foundation for developing diagnostic tests and targeted therapies, potentially offering a pathway toward improved outcomes in pancreatic cancer management.