Discovering Cell Types by Integrating Single-cell RNA-sequencing and Protein Interaction Network

Single-cell RNA-sequencing (scRNA-seq) investigates the transcriptome of individual cells, and discovering of cell types is one of the prominent job for scRNA-seq. Current algorithms address this issue by only exploiting the profiles of scRNA-seq, which is insufficient to fully characterize cell types. How to integrate the accumulated protein interaction networks and scRNA-seq is promising to identify cell types. In this study, we proposed a graph regularized nonnegative matrix factorization algorithm for the identification of cell types (called GrNMFCT), where the protein interaction network and scRNA-seq are integrated. Specifically, the profiles of scRNA-seq is factorized to obtain the latent features of cells, and protein interaction network is incorporated into the objective function via the regularization strategy. In this case, cell type discovery is formulated as an optimization problem. The experimental results demonstrate that the proposed algorithm is more accurate than state-of-the-art on the identification of cell types on both artificial and biological data. The proposed algorithm provides an effective way for the integrative analysis of scRNA-seq data.