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

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional, sparse data.
• Accurate cell type classification is crucial for understanding biological systems.
• Existing methods face challenges in efficiently processing complex scRNA-seq data.

Purpose of the Study:

• To develop a novel computational method for single-cell RNA sequencing data classification.
• To leverage graph convolutional networks (GCN) and Mamba models for improved feature extraction and efficiency.
• To enhance the accuracy and interpretability of cell type identification.

Main Methods:

• Constructing a cell graph using GCN to capture cell-cell topological relationships.
• Employing the Mamba model (selective state-space models) for efficient processing of sparse, high-dimensional gene expression data.
• Implementing an end-to-end architecture integrating GCN and Mamba for computational efficiency.

Main Results:

• scGMB achieved high classification accuracies across five diverse datasets (Zheng68K, Zhengsorted, BaronHuman, BaronMouse, AMB), reaching up to 99.5%.
• The method outperformed existing mainstream approaches in cell type identification.
• Demonstrated effective extraction of complex gene expression patterns and topological features.

Conclusions:

• scGMB provides a high-performance and stable tool for single-cell data analysis.
• The integrated GCN and Mamba approach enhances computational efficiency and result interpretability.
• Future work includes extending scGMB to spatial transcriptomics and multi-omics data analysis for broader biomedical applications.