GraphSTAR: Proximal Operator-Based Graph Neural Network Enhanced by Dynamic Graph Aggregation for Spatial...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics provides molecular profiles with spatial context, crucial for understanding biological regulation.
• Integrating spatial coordinates with high-dimensional gene expression data presents significant challenges.
• Existing methods often overlook long-range relationships in gene expression data.

Purpose of the Study:

• To introduce GraphSTAR, a novel approach for integrating spatial and gene expression data.
• To address the limitations of current methods in capturing both local and long-range relationships.
• To enhance the analysis of spatial expression patterns in transcriptomics.

Main Methods:

• GraphSTAR encodes spatial and gene expression data into undirected graphs representing local proximity and global similarity.
• A graph aggregation process integrates these into a joint graph structure.
• A reassembled graph neural network refines spatial-informed latent representations.

Main Results:

• GraphSTAR effectively models both local neighborhood relationships and long-range functional associations.
• Experiments show GraphSTAR outperforms state-of-the-art methods on benchmark datasets.
• The method demonstrates superior performance in spatial domain identification and cell-type annotation.

Conclusions:

• GraphSTAR offers a robust framework for integrating diverse spatial transcriptomics data.
• The approach enhances the deciphering of spatial gene expression patterns.
• This method advances the capabilities of spatial transcriptomics analysis.