A Novel GCN-Based Cell Annotation Method for Single-Cell RNA Sequencing Data

Single cell RNA sequencing (scRNA-seq) can capture transcriptomic features, reveal cellular heterogeneity and rare cell types at the single-cell level, making it an essential tool across multiple biomedical studies. With the exponential growth of scRNA-seq data, accurately annotating cell type has become a key challenge. Traditional methods with manual labeling struggle to handle high dimensionality and complexity of single-cell profiling. Recently, automatic annotation based on deep learning has significantly improved accuracy and efficiency, especially graph neural network shows great potential in novel type identification. However, these methods still face limitations due to the dimensionality and data noise. Here we propose a novel classification method based on graph convolutional network and Transformer. Through network augmentation, it can fully explore the complex relationships within cell populations, ensuring the credibility of the graph structure, and Transformer can extract global information and address long-distance dependencies. Our proposed method achieves superior performance on multiple real scRNA-seq datasets compared to existing approaches, with an average accuracy ranging from 92.3% to 99.8%; meanwhile, our method demonstrates robustness and generalizability, showcasing its potential in advancing single-cell type annotation in diverse biological contexts.