Identification of DNA Methylation Markers Using Feature Selection and Deep Learning

Abstract

Current tools used to gather DNA methylation data can now retrieve well over 850,000 unique methylation values per sample. As these datasets tend to be too large for a human to reasonably parse through by themselves, computational methods need to be used. Deep learning, a form of machine learning, is often used to make predictions and classifications based on data (e.g., allowing a computer to generally tell if a picture contains a cat or a dog). Previous DNA methylation research for cancer prediction has been shown to be effective. The current study utilized breast cancer methylation-array data from The Cancer Genome Atlas (TCGA), specifically the TCGA-BRCA dataset. This dataset contains over 800 unique files containing methylation data, with each file representing a single sample. Each methylation-array file can contain around 450k different methylation markers, each of which may have a unique methylation beta value (ranging from 0 to 1) for a specific sample.