Abstract:
Pancreatic ductal adenocarcinoma (PDAC) is the most common type of pancreatic cancer. Due to its high lethality and difficult diagnosis, PDAC is a prime candidate for biomarker discovery using novel approaches. This study uses high-throughput gene expression and protein-protein interaction data to apply differential gene expression analysis, constraint-based reconstruction and analysis (COBRA) of the metabolic network, and the construction of protein-protein interaction (PPI) and gene coexpression networks, comparing cancerous and healthy pancreatic tissue. Clustering of PPI and coexpression networks is used to place differential expression and COBRA results in biological context and find gene clusters which describe cancer-related processes. Differential expression analysis on TCGA and GTEx databases found 826 differentially expressed genes (DEGs). For COBRA methods, gene expression data was used to filter the 10,600 metabolic reactions in the Recon3D generic human metabolic model into cancer and healthy tissue-specific models, containing 5,879 and 5,812 reactions, respectively. The metabolic flux profiles of these models were discovered via flux sampling, revealing 1,960 significantly different internal reactions and 338 significantly different exchange reactions. These were narrowed down to 250 internal reactions and 14 exchange reactions whose mean flux difference between two models was greater than two standard deviations of the flux distributions. 55 genes were found to both be DEGs and exhibit significant flux differences in their associated reactions. These genes and the top 5 PPI or coexpression clusters with the highest proportion of DEGs or genes with significant metabolic differences were annotated with Gene Ontology (GO) and Disease Ontology (DO) terms, as well as with known biological pathways. The annotation revealed associations between cancer genes and cell-cell adhesion, immune response, and reactive oxygen species (ROS) stress processes, among others.