Unified K-means coupled self-representation and neighborhood kernel learning for clustering single-cell RNA-sequencing data.

Single-cell RNA sequencing (scRNA-seq) technology obtains transcriptomics information of cells individually, and it allows the detection of cell types and subtypes at the cellular level. Nevertheless, the large number of noise and high dimension of expression profiles in scRNA-seq data cause many problems in down-stream analysis. Subspace clustering has been widely applied to scRNA-seq data analysis and made great progress. However, it only holds for linear subspaces and carries out similarity learning and subsequent clustering separately, which can generate severe information loss. To this end, we develop a novel unified framework for learning a consensus similarity matrix and a representation matrix jointly from the scRNA-seq data. Furthermore, by learning a low-rank kernel matrix, we try to fully exploit the diversity property of the kernel matrix and allow the optimal kernel to reside in the neighborhood of the candidate kernels. In addition, to solve the non-convex resultant optimization problem, an alternating direction method with multipliers (ADMM) is designed. Experimental results on various scRNA-seq datasets (including Cardiac epidermal cells, neural cells, blood cells, etc.) validate the superiority of the proposed approach when compared with other state-of-the-art ones. [ABSTRACT FROM AUTHOR]