Abstract
The term ‘epigenetic’ refers to all heritable alterations that occur in a given gene function without having any change on the DeoxyriboNucleic Acid (DNA) sequence. Epigenetic modifications play a crucial role in development and differentiation of various diseases including cancer. The specific epigenetic alteration that has garnered a great deal of attention is DNA methylation, i.e., the addition of a methyl-group to cytosine. Recent studies have shown that different tumor types have distinct methylation profiles. Identifying idiosyncratic DNA methylation profiles of different tumor types and subtypes can provide invaluable insights for accurate diagnosis, early detection, and tailoring of the related treatment for cancer. In this study, our goal is to identify the informative genes (biomarkers) whose methylation level change correlates with a specific cancer type or subtype. To achieve this goal, we propose a novel high dimensional learning framework inspired by the dynamic data driven application systems paradigm to identify the biomarkers, determine the outlier(s) and improve the quality of the resultant disease detection. The proposed framework starts with a principal component analysis (PCA) followed by hierarchical clustering (HCL) of observations and determination of informative genes based on the HCL predictions. The capabilities and performance of the proposed framework are demonstrated using a DNA methylation dataset stored in Gene Expression Omnibus (GEO) DataSets on lung cancer. The preliminary results demonstrate that our framework outperforms the conventional clustering algorithms with embedded dimension reduction methods, in its efficiency to identify informative genes and outliers, and removal of their contaminating effects at the expense of reasonable computational cost.