ARGLRR: An Adjusted Random Walk Graph Regularization Sparse Low-Rank Representation Method for Single-Cell RNA-Sequencing Data Clustering

Abstract

Researchers may now explore biological concerns at the cell level because of the advancement of single-cell transcriptome sequencing technologies. One of the primary applications of single-cell RNA-seq (scRNA-seq) data is to identify cell types by clustering to reveal cell heterogeneity. However, due to characteristics such as higher noise and lesser coverage of scRNA-seq, the accuracy of existing clustering methods is compromised. Here, we propose a method called Adjusted Random walk Graph regularization Sparse Low-Rank Representation (ARGLRR), a practical sparse subspace clustering method, to identify cell types. The basic Low-Rank Representation (LRR) model focuses primarily on the global structure of data. We add adjusted random walk graph regularization to the framework of LRR, which makes up for the lack of local structure capture capability of LRR. With this method, the local and global structure of the scRNA-seq data will be captured. By imposing the similarity constraint on the LRR model, the cell-to-cell similarity estimation process further enhances the capacity of the proposed model to capture the global structural relationships between cells. The results on nine published scRNA-seq datasets demonstrate that ARGLRR outperforms other advanced comparison methods. Our method improves 6.99% and 5.85% over the best-performing comparison method in NMI and ARI metrics on the scRNA-seq datasets clustering experiments, respectively. We also use UMAP to visualize the learned similarity matrix and find that the similarity matrix obtained by ARGLRR improves the separation of cell types.