Clustering Single-Cell RNA Sequencing Data by Deep Learning Algorithm

Abstract

The development of single-cell RNA sequencing (scRNA-seq) technology provides a good opportunity to study cell heterogeneity and diversity. Especially, clustering is an important step in scRNA-seq analysis. With the advance of technology, many scRNA-seq data are available, which develop a lot of clustering methods. However, the existing methods usually employ the gene expression data, ignoring the related information between genes and the structure information in data. Therefore, we propose a new method (NDMgcn) to reconstruct the gene expression data based on the association of gene network, and cluster the data by Variational Autoencoder (V AE) and Graph Convolutional Network (GCN). The V AE learns low-dimensional information and the GCN learns structural information. The experimental results indicate that NDMgcn outperforms other popular algorithms in terms of NMI and ARI metrics. It provides a new insight for clustering scRNA-seq data from the network perspective.