scAFGCC: An augmentation-free graph contrastive clustering method for scRNA-seq data analysis

The emergence of single-cell RNA sequencing (scRNA-seq) has provided researchers with a powerful tool to investigate cell heterogeneity and human diseases at the level of individual cells. Cell clustering is a crucial step in scRNA-seq data analysis to identify marker genes and recognize cell types. However, scRNA-seq data present challenges for clustering tasks due to their high dimensionality, sparsity, and noise. Although some contrastive learning methods have achieved good results in clustering scRNA-seq data, they are highly sensitive to data augmentation schemes. Here, we propose scAFGCC, a novel augmentation-free graph contrastive clustering method that combines graph convolutional network (GCN) and contrastive learning to exploit inter-cell relationships. scAFGCC does not require data augmentations or negative samples to learn graph representations. Instead, we generate positive samples by exploring the local structural information and the global semantics of the target nodes. We integrate feature representation learning with clustering tasks. Additionally, we introduce a reconstruction module that pretrains the model, facilitating faster training and improved performance. Our experiments on 24 simulated and 13 real datasets show that scAFGCC outperforms seven state-of-the-art methods in terms of accuracy and robustness. We also apply scAFGCC to downstream tasks such as cell annotation and marker gene identification.