Prognostic prediction of a 12-methylation gene-based risk score system on pancreatic adenocarcinoma.

Pancreatic adenocarcinoma (PAAD) accounts for ~85% of all pancreatic cancer cases and is associated with a less favorable prognosis. Aberrant DNA methylation may influence the progression of PAAD by inducing abnormal gene expression. Methylation data of PAAD samples with prognosis information were obtained from The Cancer Genome Atlas (training set) and European Bioinformatics Institute Array Express databases (validation sets). Using the limma package, the differentially methylated genes in the training dataset were screened. Combined with the Weighted Gene Co-expression Network Analysis package, the co-methylated genes in key modules were identified. Then, a cor.test function in R software was applied to explore the functions of key the methylated genes. Correlation analyses of the expression levels and methylation levels of key methylated genes were performed, followed by identification of methylated genes associated with prognosis using Univariate Cox regression analysis. The optimal combination of prognosis related methylated genes was determined using a Cox-Proportional Hazards (Cox-PH) model. Subsequently, the risk score prognostic prediction system was constructed by combining the Cox-PH prognosis coefficients of the selected optimized genes. Based on the constructed risk score system, samples in all datasets were divided into high and low risk samples and the survival status was compared using survival curves. Furthermore, the correlation between independent prognostic factors and the risk score system was determined using the survival package. A total of 50 genes associated with prognosis of PAAD and a 12-gene optimal combination were obtained, including: CCAAT/enhancer binding protein α, histone cluster 1 H4E, STAM binding protein-like 1, phospholipase D3, centrosomal protein 55, ssDNA binding protein 4, glutamate AMPA receptor subunit 1, switch-associated protein 70, adenylate-cyclase activating polypeptide 1 receptor 1, yippee-like 3, homeobox C4 and insulin-like growth factor binding protein 1. Subsequently, a risk score prognostic prediction system of these 12 genes was constructed and validated. In addition, pathological N category, radiotherapy and risk status were identified as independent prognostic factors. Overall, the risk score prognostic prediction system constructed in the present study may be effective for predicting the prognosis of patients with PAAD.