Single Cell Clustering with Sparse Similarity Matrix Learning

Abstract

Listen

Translate article

Along with the development of single-cell RNA sequencing (scRNA-seq) techniques, there is an unparalleled resolution to explore cellular heterogeneity by identifying cell types. A number of studies have been proposed on cell type identification, many of which rely on the cell-cell similarity construction. In this study, we present a single-cell clustering method based on learning sparse similarity matrices (LSSM), to identify cell types more accurately. LSSM is a novel analysis framework considering a more number of similar cells to construct sparse cell-cell similarity matrices with sparse subspace theory. We apply LSSM to eight single-cell RNA sequencing data sets and compare its performance with several state-of-the-art methods. The results show that LSSM is superior to most competing methods on multiple data sets, which demonstrates the potential usefulness of our proposed method.