SpaceBF: spatial coexpression analysis using Bayesian fused approaches in spatial omics datasets.

Area of Science:

• Spatial omics
• Molecular biology
• Bioinformatics

Background:

• Spatial omics technologies allow molecular measurements across tissue locations.
• Identifying co-expressed molecule pairs in spatial data is challenging.
• Existing methods lack robustness for spatial co-expression analysis.

Purpose of the Study:

• Introduce SpaceBF, a Bayesian fused modeling framework.
• Estimate local and global co-expression of molecules in spatial omics data.
• Enhance the analysis of spatial molecular interactions.

Main Methods:

• Developed a Bayesian fused modeling framework (SpaceBF).
• Utilized a fused horseshoe prior for spatial smoothness on an adjacency graph.
• Benchmarked against geospatial metrics (Moran's I, Lee's L) and alternative priors (ICAR, Matérn).

Main Results:

• SpaceBF demonstrates higher specificity and power in simulations compared to existing methods.
• The framework effectively estimates both local and global co-expression levels.
• Applied to spatial transcriptomics and proteomics, SpaceBF identified cancer-relevant interactions and cell-cell communication patterns.

Conclusions:

• SpaceBF provides a principled and uncertainty-aware approach for spatial omics co-expression analysis.
• The method enhances the discovery of molecular interactions and biological pathways in tissues.
• SpaceBF is valuable for understanding complex spatial omics data, including ligand-receptor signaling.