Adaptive Total Variation Constraint Hypergraph Regularized NMF and Its Application on Single-Cell RNA-Seq Data

Single-cell RNA sequencing (scRNA-seq) data provides a whole new view to study the development of disease and cell differentiation. As the explosive increment of scRNA-seq data, effective models are demanded to mining the intrinsic biological information. In this paper, we propose a novel nonnegative matrix factorization (NMF) method for clustering and gene co-expression network analysis, termed Adaptive Total Variation Constraint Hypergraph Regularized NMF (ATV-HNMF). Based on the gradient information, ATV-HNMF can adaptively select the different schemes to denoise in the cluster or preserve the cluster boundary information between clusters. Besides, ATV-HNMF incorporates hypergraph regularization which can consider high-order relationships between cells, which is helpful to reserve the intrinsic structure of the original data space. Experiments show that the performance on clustering outperforms other compared methods. And the related genes mined by co-expression network construction are consistent with the previous research, which illustrates that our model is effective and useful.