Deciphering spatial domains from spatial multi-omics with SpatialGlue

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Spatial omics technologies enable simultaneous data acquisition from single tissue sections.
• Integrating multi-omics data spatially is crucial for understanding tissue architecture and function.
• Existing methods lack the ability to fully leverage spatially resolved multi-omics information.

Purpose of the Study:

• To introduce SpatialGlue, a graph neural network model for integrating spatial multi-omics data.
• To decipher spatial domains and identify cell types by combining different omics measurements.
• To provide a scalable tool for holistic analysis of cellular and tissue properties.

Main Methods:

• Developed SpatialGlue, a graph neural network with a dual-attention mechanism.
• Performed intra-omics integration of spatial location and omics measurements.
• Conducted cross-omics integration for enhanced spatial domain deciphering.
• Applied the method to spatial epigenome-transcriptome and transcriptome-proteome data.

Main Results:

• SpatialGlue accurately resolved spatial domains, including brain cortex layers.
• The method identified previously unannotated spleen macrophage subsets in distinct zones.
• SpatialGlue demonstrated superior performance in capturing anatomical details compared to existing methods.
• The tool successfully integrated three omics modalities and scaled well with data size.

Conclusions:

• SpatialGlue effectively integrates spatial multi-omics data, revealing intricate tissue structures.
• The model enhances the understanding of cellular heterogeneity and spatial organization.
• SpatialGlue offers a powerful approach for comprehensive spatial multi-omics analysis.