Abstract
BACKGROUND: Genomic analysis is the promising tool to clear understanding of the tumorigenesis and guide molecular classification for pancreatic cancer. Our purpose was to develop a critical predictive model for prognosis in pancreatic carcinoma, based on the genomic data. METHODS: The online The Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC) datasets were queried as training and validation cohorts for comprehensive bioinformatic analysis. We applied Lasso and multivariate Cox regression to shrink genes and construct predictive model. RESULTS: A four genes model (DNAH10: HR = 0.71, 95% CI = 0.57-0.88, HSBP1L1: HR = 1.51, 95% CI = 1.18-1.92, KIAA0513: HR = 0.69, 95% CI = 0.50-0.96, and MRPL3: HR = 3.73, 95% CI = 2.03-6.86), was proposed and validated. The C-index was 0.73 (95% CI: 0.7-0.77). Patients in high-risk and low-risk group, stratified by model, suffered significantly different overall survival time (15.1 vs. 49.3 months, p < 0.0001 in TCGA; 423 vs. 618 days, p = 0.038 in ICGC). Taken clinical parameters into consideration, the risk-score was independent marker in clinical subpopulation. To explore the molecular mechanisms, 579 differential expression genes (DEG) in two groups were identified by edgeR. Functional enrichment of DEG indicated neuro-endocrine activity was the potential mechanism for the discrepant prognosis. CONCLUSION: A specific four genes signature with the ability to predicted survival of pancreatic carcinoma was generated, which may indicate the connection between neuro-endocrine activity and patients' prognosis.