Annotation of spatially resolved single-cell data with STELLAR.

Area of Science:

• Spatial biology
• Computational biology
• Single-cell analysis

Background:

• Accurate cell-type annotation is vital for understanding tissue organization in spatial biology.
• Existing computational methods for spatial single-cell data often neglect spatial organization, relying on techniques for dissociated cells.

Purpose of the Study:

• To introduce STELLAR, a novel geometric deep learning method for cell-type discovery and identification in spatially resolved single-cell datasets.
• To develop a computational tool that leverages spatial organization for improved cell annotation.

Main Methods:

• STELLAR employs geometric deep learning to analyze spatially resolved single-cell data.
• The method learns cell representations that capture higher-order tissue structures.
• It automatically assigns cells to known types and discovers novel cell types and states.

Main Results:

• STELLAR successfully annotates cell types by incorporating spatial organization.
• The method demonstrates the ability to transfer annotations across different tissues and donors.
• Applied to CODEX and multiplexed RNA imaging data, STELLAR achieved significant time savings in annotating millions of cells.

Conclusions:

• STELLAR offers an advanced computational approach for cell-type annotation in spatial biology.
• By integrating spatial information, STELLAR improves the accuracy and efficiency of cell identification.
• This method has broad applications in large-scale spatial atlases and biological research.