Inferring spatial single-cell-level interactions through interpreting cell state and niche correlations learned by...

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Cell-cell interactions (CCI) are crucial for tissue development and function.
• Spatial transcriptomics (ST) enables single-cell resolution analysis of CCI but faces challenges in interpretability and spatial encoding.
• Existing methods struggle with limited ligand-receptor data and understanding complex cellular communication networks.

Purpose of the Study:

• To introduce GITIII, a self-supervised graph transformer model for inferring and interpreting spatial cell-cell interactions.
• To address limitations in current ST analysis, including insufficient spatial encoding and interpretability.
• To provide a framework for understanding how cellular neighborhoods influence gene expression and cell states.

Main Methods:

• Developed GITIII, a lightweight, interpretable, self-supervised graph transformer model.
• Conceptualized cells as words and their neighborhood as context to infer CCI.
• Analyzed gene expression correlations between cell states and their microenvironment (niche).
• Applied GITIII to four diverse ST datasets across species, organs, and platforms.

Main Results:

• GITIII successfully inferred and statistically interpreted spatial CCI patterns.
• Enabled visualization of spatial CCI patterns and construction of CCI networks.
• Facilitated CCI-informed cell clustering and understanding of sender-receiver cell influence.
• Demonstrated effectiveness in brain and tumor microenvironments.

Conclusions:

• GITIII provides a powerful and interpretable tool for analyzing spatial cell-cell interactions from ST data.
• The model advances our ability to decipher complex cellular communication in various biological contexts.
• GITIII opens new avenues for understanding tissue development, organ function, and disease mechanisms through spatial biology.