A Machine Learning Approach Towards Prediction of Pancreatic Cancer Using Gene Expression and DNA Methylation

Abstract

DNA methylation is a process that can affect gene accessibility and therefore gene expression. Methylation can affect genes that are associated with suppressing or contributing to tumor growth and progression. In this research, we examined the potential of developing a scalable feature selection and deep learning framework capable of processing high dimensional genomic datasets to identify methylation sites in the human genome responsible for pancreatic ductal adenocarcinoma (PDAC). The methylation data being analyzed consisted of almost 485,578 CPG markers. Feature selection using Random Forest and Anova was performed along with random undersampling. Selected markers and gene expression data was analyzed to identify enriched pathways. Genes identified in pathways were studied for tumor progression. When compared with a list of 339 genes that were identified from literature, our results returned 98 common tumor genes including “KRAS", "BRCA1", "BRCA2", "PALB2", "CDKN2A", "TP53", "SMAD4". This reveals that the stratified feature selection technique is indeed useful at identifying important features. Finally a deep learning classifier was developed for identifying patients with PDAC. Data for this research was obtained from TCGA-PAAD project. Current work is focused on analyzing gene expression data and combine outcomes with methylation dataset to get deeper insights into the disease mechanisms. PDAC, or pancreatic ductal adenocarcinoma, is the most prevalent kind of pancreatic cancer, making up over 90% of cases. Aggressive tumor growth, early metastasis, and resistance to therapy are characteristics of this highly fatal cancer. Improving diagnosis, prognosis, and treatment outcomes require a thorough understanding of the molecular pathways underlying PDAC development and progression. The proposed research outcomes could aid in the development of non-invasive diagnostic tests and personalized treatment strategies.