Abstract: Single-cell RNA sequencing (scRNA-seq) provides high-resolution observation tools at the cell level for biological domains, such as embryonic development, cancer evolution and cell differentiation. A key step in using scRNA-seq data is to cluster cells with similar biological functions into one group. However, the current clustering methods are not able to perform the clustering task well in a large number of high-dimensional and complex scRNA-seq data, and don’t use the structural relationship information between samples. Here, we propose a GCN based deep clustering framework, named Self-supervised Low-pass Filted Graph Clustering Networks (SLFGCN). Firstly, a new propagation method of graph convolutional network is proposed. For the proposed method, the graph information in the spectral domain passes through the frequency response function of the low-pass filter to obtain smoother node feature representation, which is more conducive to the clustering task. Secondly, we use the self-supervised module to optimize the network based on the representation learned from the low-pass filted GCN module and the representation learned from the graph auto-encoders module, which can obtain better clustering effect. Experiments indicate that our model outperforms the state-of-the-art methods in various evaluation metrics on real datasets. Further, the visualization results show that our model provides representations generating better intra-cluster compactness and inter-cluster separability.