[Article] Deriving disease modules from the compressed transcriptional space embedded in a deep autoencoder.
Summary: Using deep neural networks(DNN) to classify diseases using genetic data is popular these days. In this article, they hypothesized that disease-relevant modules of genes can be discovered within the autoencoder (AE) representations. They compared shallow and deep AE with various node sizes, and showed deep AE works better. Also, they also showed that each different layer captures gradients of biology. By overlapping top 1000 genes for each disease with GWAS, they found a highly significant association for at least one layer in all tested diseases.