stMCFN: A Multi-view Contrastive Fusion Method for Spatial Domain Identification in Spatial Transcriptomics

Rapid advances in spatial transcriptomics allow for sequencing gene expression profiles while preserving spatial information. Many spatial clustering methods based on graph neural networks have been proposed, but they often cannot adaptively learn the complex relationships between gene expression and spatial information, which makes it challenging to identify spatial domains effectively. In this paper, we propose stMCFN, a multi-view contrastive fusion network based on graph autoencoders for spatial domain identification. To fully mine the potential structural information in spatial transcribed data, multiple views are constructed using gene expression and spatial information for self-supervised contrastive learning of graph autoencoders without data enhancement. This method generates shared embedding from different views and obtains clustering-friendly feature representation through an attention-based feature fusion module to adaptively integrate embedding from different views. Finally, the experimental results on the human dorsolateral prefrontal cortex dataset and the mouse brain anterior dataset show that stMCFN's ability to identify spatial domains is significantly superior to other state-of-the-art methods. The code is available at https://github.com/JING-ING/stMCFN.