Exploring DNA Methylation Data of Lung Cancer Samples with Variational Autoencoders

摘要

Lung cancer causes over one million deaths each year worldwide. DNA methylation is a well-defined epigenetics factor in genome data analyses for model training. In this article, we explore the applications of unsupervised deep learning method, variational autoencoders, using DNA methylation data of lung cancer samples downloaded from the GDC TCGA project and perform further work with latent features. We show the logistic regression classifier on the encoded latent features accurately classifies cancer subtypes.