scSFCL:Deep clustering of scRNA-seq data with subspace feature confidence learning

Abstract

The rapid development of single-cell RNA sequencing(scRNA-seq) technology has spawned a variety of single-cell clustering methods. These methods combine statistics and bioinformatics to reveal differences in gene expression between cells and the diversity of cell types. Deep exploration of single-cell data is more challenging due to the high dimensionality, sparsity and noise of scRNA-seq data. Discriminative attribute information is often difficult to be fully utilised, while traditional clustering methods may not accurately capture the diversity of cell types. Therefore, a deep clustering method is proposed for scRNA-seq data based on subspace feature confidence learning called scSFCL. By dividing the subspace based on kernel density, discriminative feature subsets are filtered. The feature confidence of the subset is learned by combining the graph convolutional network (GCN) with weighting. Also, scSFCL facilitates the complementary fusion of generic structural and idiosyncratic information through a mutually supervised clustering that integrates GCN and a denoising variational autoencoder based on zero-inflated negative binomials (DVAE-ZINB). By validation on multiple scRNA-seq datasets, it is shown that the clustering performance of scSFCL is significantly improved compared with traditional methods, providing an effective solution for deep clustering of scRNA-seq data.