Spatial domain identification method based on multi-view graph convolutional network and contrastive learning

Spatial transcriptomics is a rapidly developing field of single-cell genomics that quantitatively measures gene expression while providing spatial information within tissues. A key challenge in spatial transcriptomics is identifying spatially structured domains, which involves analyzing transcriptomic data to find clusters of cells with similar expression patterns and their spatial distribution. To address these challenges, we propose a novel deep-learning method called DMGCN for domain identification. The process begins with preprocessing that constructs two types of graphs: a spatial graph based on Euclidean distance and a feature graph based on Cosine distance. These graphs represent spatial positions and gene expressions, respectively. The embeddings of both graphs are generated using a multi-view graph convolutional encoder with an attention mechanism, enabling separate and co-convolution of the graphs, as well as corrupted feature convolution for contrastive learning. Finally, a fully connected network (FCN) decoder is employed to generate domain labels and reconstruct gene expressions for downstream analysis. Experimental results demonstrate that DMGCN consistently outperforms state-of-the-art methods in various tasks, including spatial clustering, trajectory inference, and gene expression broadcasting.Author summary: Spatial transcriptomics technology not only provides high-resolution gene expression data but also completely preserves the spatial location information of each sequencing spot in tissue sections, offering an unprecedented multi-dimensional perspective for in-depth exploration of tissue heterogeneity and dissection of cellular microenvironments and functional partitions. Similar to the cell clustering task in single-cell data analysis, one of the core challenges in spatial transcriptomics data analysis is spatial domain identification—using algorithms to cluster sequencing spots with adjacent spatial locations and similar gene expression patterns into biologically meaningful functional regions. Although existing methods can achieve preliminary spatial domain partitioning, accurately capturing global semantic associations and balancing the modeling capabilities of local features and global structures remain critical scientific challenges when faced with non-linear gene expression patterns and spatial distribution relationships in complex tissues. Here, we propose a novel algorithmic framework that integrates multi-view graph convolutional networks and contrastive learning. Results across multiple datasets generated by different technologies demonstrate that our method significantly enhances the accuracy and biological interpretability of spatial domain identification.