Single-Cell Transcriptome Analysis in Melanoma Using Network Embedding

Abstract

Single-cell sequencing technology provides insights into the pathology of complex diseases like cancer. Here, we proposed a novel computational framework to explore the molecular mechanisms of cancer called melanoma. We first constructed a disease-specific cell–cell interaction network after data preprocessing and dimensionality reduction. Second, the features of cells in the cell–cell interaction network were learned by node2vec which is a graph embedding technology proposed previously. Then, consensus clusters were identified by considering different clustering algorithms. Finally, cell markers and cancer-related genes were further analyzed by integrating gene regulation pairs. We exploited our model on two independent datasets, which showed interesting results that the differences between clusters obtained by consensus clustering based on network embedding (CCNE) were observed obviously through visualization. For the KEGG pathway analysis of clusters, we found that all clusters are extremely related to MicroRNAs in cancer and HTLV-I infection, and the hub genes in cluster specific regulatory networks, i.e., ETS1, TP53, E2F1, and GATA3 are highly associated with melanoma. Furthermore, our method can also be extended to other scRNA-seq data.