Detecting cell type from single cell RNA sequencing based on deep bi-stochastic graph regularized matrix factorization

Abstract

The application of fruitful achievement of single-cell RNA-sequencing (scRNA-seq) technology has generated huge amount of gene transcriptome data. It has provided a whole new perspective to analyze the transcriptome at single-cell level. Cluster analysis of scRNA-seq is an efficient approach to reveal unknown heterogeneity and functional diversity of cell populations, which could further assist researchers to explore pathogenesis and biomarkers of diseases. In this paper, we propose a new cluster method (DSINMF) based on deep matrix factorization to detect cell type in the scRNA-seq data. In our method, the feature selection is used to reduce redundant features. Then, the imputation method is utilized to impute dropout events. Further, the dimension reduction is utilized to reduce the impact of noise. Finally, the deep matrix factorization with bi-stochastic graph regularization is employed to cluster scRNA-seq data. To evaluate the performance of DSINMF, eight datasets are used as test sets in the experiment. The experimental results show DSINMF outperformances than other state-of-the-art methods in clustering performance.