Decoding spatial transcriptomics across multicellular and subcellular resolutions.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Current whole-genome spatial transcriptomics (ST) platforms struggle to resolve gene expression at the single-cell level.
• Existing computational methods primarily analyze data at the spot level, hindering biological interpretation.
• Reconstructing transcriptomes at single-cell resolution is crucial for understanding cellular heterogeneity and function within tissues.

Purpose of the Study:

• To introduce STARS (Spatial Transcriptomics across Resolutions for Single Cells), a novel computational method.
• To enable accurate reconstruction of single-cell gene expression from various ST platforms.
• To advance the downstream analysis of spatial transcriptomics across different resolutions.

Main Methods:

• Leveraging a Vision Transformer model and contrastive learning.
• Integrating high-resolution histology images with spot-level transcriptomics data.
• Applying the STARS method across tissue, individual cell, and molecular levels.

Main Results:

• STARS successfully reconstructs single-cell gene expression from multicellular and subcellular ST platforms.
• The method identifies specific tissue structures, immune regions, and diverse immune cell subtypes (e.g., CD4/CD8 T cells, CAFs, macrophages).
• STARS reveals tertiary lymphoid structures in cancer and detects immune cell population shifts in response to infection, with improved cell type separation and marker identification.

Conclusions:

• STARS provides biologically relevant insights into tissue architecture and gene expression at the single-cell level.
• The method enhances the downstream analysis of spatial transcriptomics, offering greater accuracy and resolution.
• STARS represents a significant advancement for understanding complex biological systems through spatial transcriptomics.