stDGCN: A dual-augmentation graph convolutional network for identifying spatial domains with attention mechanism.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics enables high-resolution gene expression analysis within tissue context.
• Accurate identification of spatial domains is hindered by data sparsity and noise.

Purpose of the Study:

• To develop a robust framework for spatial domain identification in spatial transcriptomics data.
• To address challenges posed by data sparsity and noise using advanced deep learning techniques.

Main Methods:

• Proposed a dual-augmentation graph convolutional network (stDGCN) incorporating spatial neighborhood enhancement and Gaussian perturbation.
• Employed an attention-based fusion module and consistency regularization for improved representation stability.
• Utilized a zero-inflated negative binomial decoder for gene expression profile reconstruction.

Main Results:

• stDGCN demonstrated superior performance in spatial domain identification across six diverse spatial transcriptomics datasets.
• Achieved higher clustering accuracy and robustness compared to seven state-of-the-art methods.
• Validated on human brain, breast cancer, and mouse datasets.

Conclusions:

• stDGCN offers a robust and interpretable solution for spatial domain analysis in transcriptomics.
• The framework enhances the understanding of tissue architecture and gene expression patterns.
• Potential for significant applications in downstream biological and clinical research.