SCDMSC: Deep Multi-View Subspace Clustering for Single-Cell Multi-Omics Data

With the rapid development of single-cell sequencing technologies, various clustering methods have emerged, significantly enhancing our understanding of cellular heterogeneity and biological processes. However, many existing clustering methods are primarily designed for single-cell RNA sequencing data, mono-omic data typically only capture information from a specific aspect, failing to comprehensively reveal the diversity of cells. The emergence of single-cell multi-omics sequencing technologies facilitates better exploration of cell diversity, but it has also posed challenges. How to effectively utilize the available information across multiple omics data and explore the global structure of cells has become a new challenge. To address this issue, we propose a deep multi-view subspace clustering (scDMSC) model to identify cell types from single-cell multi-omics data, which can learn both cell representations and subspaces simultaneously. To explore the global structure of cells, scDMSC introduces a novel subspace affinity. Furthermore, a collaborative training strategy is used to learn the consensus and complementary information of single-cell multi-omics data. The superiority of scDMSC is demonstrated through experimental results on three diverse single-cell multi-omics datasets.