Discovery and generalization of tissue structures from spatial omics data

Area of Science:

• Spatial biology
• Computational pathology
• Bioinformatics

Background:

• Tissues exhibit complex organization across multiple scales.
• In situ molecular profiling technologies advance understanding of tissue structure-function relationships.
• Consistent identification of functional units across datasets is challenging, often requiring manual annotation.

Purpose of the Study:

• To present spatial cellular graph partitioning (SCGP), a flexible, unsupervised method for annotating tissue structures.
• To introduce SCGP-Extension for generalizing tissue structure labels to new samples via data integration and discovery.
• To enable robust and accurate identification of key functional units in spatial datasets.

Main Methods:

• Spatial cellular graph partitioning (SCGP) for unsupervised annotation of tissue structures.
• SCGP-Extension pipeline for generalizing labels to unseen samples.
• Application across diverse tissue types and disease contexts.

Main Results:

• SCGP demonstrates reliable and robust partitioning of spatial data.
• Achieved best-in-class accuracy in identifying expertly annotated tissue structures.
• Downstream analysis revealed disease-relevant insights in kidney, skin, and neoplastic diseases.

Conclusions:

• SCGP provides a powerful tool for unsupervised annotation of spatial tissue data.
• SCGP-Extension facilitates data integration and discovery across experiments.
• The method has the potential to drive significant biological insight and discovery from spatial datasets.