Joint Bayesian estimation of cell dependence and gene associations in spatially resolved transcriptomic data

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics technologies allow gene expression measurement at the single-cell level with spatial localization.
• Spatial clustering reveals tissue functional organization but often involves dimension reduction, losing gene co-expression patterns.
• Existing methods may compromise spatial clustering performance by neglecting gene dependencies.

Purpose of the Study:

• To develop a joint Bayesian approach for simultaneously estimating gene and spatial cell correlations in spatial transcriptomics data.
• To preserve inherent gene co-expression patterns and cell spatial dependencies lost in dimension reduction techniques.
• To provide robust data summaries for downstream analyses in spatial transcriptomics.

Main Methods:

• Utilized matrix-variate gene expression data with spatial coordinates from single cells.
• Proposed a joint Bayesian framework to model row (gene) and column (cell) covariances.
• Applied the method to simulations and multiple real spatial transcriptomics datasets.

Main Results:

• Successfully elucidated gene co-expression networks and distinct spatial clustering patterns of cells.
• Demonstrated the preservation of gene expression dependencies and spatial cell dependencies.
• Validated the method's efficacy on both simulated and real-world biological data.

Conclusions:

• The proposed joint Bayesian approach effectively captures gene and spatial dependencies in transcriptomics data.
• This method enhances the understanding of tissue functional organization and gene regulatory networks.
• Downstream spatial-differential analysis using these estimates can facilitate the discovery of novel cell types.