Reconstructing Spatial Transcriptomics at the Single-cell Resolution with BayesDeep.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomics (SRT) techniques offer molecular profiling with spatial context.
• Current SRT methods often capture limited spatial domains, averaging gene expression over many cells.
• There is a critical need for single-cell resolution SRT data to understand tissue biology.

Approach:

• Introduced BayesDeep, a Bayesian hierarchical model for single-cell resolution SRT data reconstruction.
• Leveraged cellular morphology from histology images alongside SRT data.
• Modeled SRT count data using negative binomial regression with cell type and nuclear shape as predictors.

Key Points:

• Integrated a feature selection scheme to link morphological and molecular profiles, enhancing model robustness.
• Successfully reconstructed single-cell resolution SRT data on two real datasets.
• Demonstrated improved downstream analyses, including pseudotime and cell-cell communication.

Conclusions:

• BayesDeep enables high-resolution molecular mapping within tissues.
• Reconstructed SRT data provides deeper biological insights.
• This method advances the utility of SRT in understanding complex biological systems.