Structural-information guided fusion for spatial domain identification from spatial transcriptomics

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Accurate spatial domain identification is crucial for understanding tissue organization and disease mechanisms in spatial transcriptomics.
• Current methods often overlook intercellular interactions and struggle to preserve both local and global structural information.

Purpose of the Study:

• To develop a novel framework, SGFST (Structural-information Guided Fusion for spatial domain identification), for improved spatial domain identification in spatial transcriptomics.
• To address limitations of existing methods by integrating spatial, functional, and structural information.

Main Methods:

• SGFST integrates spatial and signal graphs using a dual-branch graph convolutional network with attention-based fusion.
• It employs a joint optimization strategy including Bayesian personalized ranking loss, zero-inflated negative binomial loss, and distance structural information constraints.
• This approach preserves local neighborhood continuity, reconstructs expression signals, and maintains global topological consistency.

Main Results:

• SGFST effectively captures complementary spatial and functional information for accurate domain identification.
• Experimental results on multiple datasets show SGFST outperforms several state-of-the-art methods.
• The framework successfully preserves local continuity, reconstructs signals, and maintains global topological structure.

Conclusions:

• SGFST provides a robust and accurate framework for spatial domain identification in spatial transcriptomics.
• The method's ability to integrate diverse information sources and preserve structural integrity offers significant advantages.
• SGFST represents a advancement in analyzing complex spatial transcriptomic data.