Bioinformatics-based construction of prognosis-related methylation prediction model for pancreatic cancer patients and its application value

Objective: Pancreatic adenocarcinoma (PAAD) is a highly malignant gastrointestinal tumor with almost similar morbidity and mortality. In this study, based on bioinformatics, we investigated the role of gene methylation in PAAD, evaluated relevant factors affecting patient prognosis, screened potential anti-cancer small molecule drugs, and constructed a prediction model to assess the prognosis of PAAD. Methods: Clinical and genomic data of PAAD were collected from the Tumor Genome Atlas Project (TCGA) database and gene expression profiles were obtained from the GTEX database. Analysis of differentially methylated genes (DMGs) and significantly differentially expressed genes (DEGs) was performed on tumorous samples with KRAS wild-type and normal samples using the “limma” package and combined analysis. We selected factors significantly associated with survival from the significantly differentially methylated and expressed genes (DMEGs), and their fitting into a relatively streamlined prognostic model was validated separately from the internal training and test sets and the external ICGC database to show the robustness of the model. Results: In the TCGA database, 2,630 DMGs were identified, with the largest gap between DMGs in the gene body and TSS200 region. 318 DEGs were screened, and the enrichment analysis of DMGs and DEGs was taken to intersect DMEGs, showing that the DMEGs were mainly related to Olfactory transduction, natural killer cell mediated cytotoxicity pathway, and Cytokine -cytokine receptor interaction. DMEGs were able to distinguish well between PAAD and paraneoplastic tissues. Through techniques such as drug database and molecular docking, we screened a total of 10 potential oncogenic small molecule compounds, among which felbamate was the most likely target drug for PAAD. We constructed a risk model through combining three DMEGs (S100P, LY6D, and WFDC13) with clinical factors significantly associated with prognosis, and confirmed the model robustness using external and internal validation. Conclusion: The classification model based on DMEGs was able to accurately separate normal samples from tumor samples and find potential anti-PAAD drugs by performing gene-drug interactions on DrugBank.