Identifying Spatial Domains From Spatial Multi-Omics Data With Graph Mutual Information and Deep Subspace Learning.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Spatial omics technologies offer insights into tissue function by linking molecular data with spatial locations.
• Identifying distinct spatial domains within tissues is crucial for understanding cellular organization and function.
• Current methods struggle with multi-omics spatial data, often neglecting spatial context or single-omics limitations.

Purpose of the Study:

• To develop a computational framework, SIMID, for accurate spatial domain identification from spatial multi-omics data.
• To integrate heterogeneous molecular profiles with spatial information for robust tissue segmentation.
• To overcome limitations of existing methods in handling complex spatial multi-omics datasets.

Main Methods:

• SIMID utilizes a graph mutual information encoder to capture spatial proximity and molecular similarity, generating omics-specific cell embeddings.
• Deep subspace learning constructs a homogeneous cell multi-layer network from heterogeneous multi-omics data.
• Low-rank and discriminative constraints are applied to decompose the network for effective domain identification.

Main Results:

• SIMID successfully integrates spatial information and multiple molecular profiles for spatial domain identification.
• Experimental results on simulated and real-world datasets show SIMID outperforms existing methods.
• The framework accurately reveals functionally distinct spatial domains within tissues.

Conclusions:

• SIMID provides an effective strategy for spatial domain identification in spatial multi-omics analysis.
• The method demonstrates superior performance by leveraging both molecular and spatial data.
• SIMID advances the analysis of spatial multi-omics data for biological discovery.