scGMB: A scRNA-seq Cell Classification Method Combining GCN and Mamba

This study proposes a single-cell RNA sequencing data classification method called scGMB, which is based on graph convolutional networks (GCN) and the Mamba model. This method constructs a cell graph to capture the topological relationships between cells and combines selective state-space models (SSMs) to extract complex features of gene expression patterns. The innovations of scGMB are as follows: (1) the GCN module captures the topological relationships between cells; (2) the Mamba module efficiently processes sparse and high-dimensional data; (3) the end-to-end architecture improves computational efficiency and interpretability of the results. We validated scGMB on five datasets: Zheng68 K, Zhengsorted, BaronHuman, BaronMouse and AMB, achieving classification accuracies of 72.2%, 91.1%, 98.9%, 99.2% and 99.5%, respectively, outperforming existing mainstream methods. Experimental results demonstrate that scGMB can effectively identify cell types, offering a high-performance and stable new tool for single-cell data analysis. Future work will expand scGMB to spatial transcriptomics and multi-omics data analysis, further optimising model performance to aid biomedical research and cell type discovery.