scGTN Improves Single-Cell RNA Sequencing Clustering with Graph Transformers

A novel framework called scGTN (deep Siamese Graph Transformer Network) has been introduced to enhance clustering in single-cell RNA sequencing (scRNA-seq) data analysis. Current methods often struggle with the inherent sparsity, noise, and complex intercellular structural information present in scRNA-seq data. scGTN addresses these challenges by formulating scRNA-seq data as a graph. It constructs two augmented graph views that act as dual perspectives, capturing complementary intercellular information. A Siamese graph transformer network is then employed to explicitly incorporate shortest-path information and node-wise distances, thereby capturing richer structural relationships between cells. Finally, an optimal transport strategy guides the cell clustering in a self-supervised manner. Extensive experiments on multiple benchmark scRNA-seq datasets demonstrate that scGTN consistently outperforms existing methods, offering a more robust and accurate approach to identifying cell types and understanding cellular heterogeneity.