A Novel Clustering-Framework of Gene Expression Data Based on the Combination Between Deep Learning and Self-organizing Map

摘要

Learning latent feature representation embedding in high-dimensional gene expression data is a crucial step for gene clustering application. Our clustering-framework method, incorporating Variational Autoencoders (VAE) into Self-Organizing Map (SOM), not only clustered gene expression data precisely, but also reduced the dimensionality of raw data effectively without any prior knowledge. The clustering results obtained from this method based on four gene datasets exhibited an impressive performance in efficiency and accuracy.