SORBET: Automated cell-neighborhood analysis of spatial transcriptomics or proteomics for interpretable sample...

Area of Science:

• Computational Biology
• Genomics
• Proteomics

Background:

• Spatially resolved transcriptomics and proteomics offer insights into biological processes.
• Analyzing spatial omics data, linking spatial information, markers, and phenotypes, is challenging.

Purpose of the Study:

• To develop a novel deep learning framework for analyzing spatial omics data.
• To predict tissue phenotypes, such as immunotherapy response, using spatial omics data.
• To understand the biological mechanisms underlying observed phenotypes.

Main Methods:

• Development of SORBET, a deep learning framework utilizing graph neural networks (GNN).
• Application of SORBET to spatial proteomics and transcriptomics data.
• Prediction of tissue phenotypes and identification of biological processes.

Main Results:

• SORBET accurately learns biologically meaningful relationships across diverse tissue structures and data acquisition methods.
• The framework successfully predicts tissue phenotypes, including response to immunotherapy.
• SORBET facilitates the understanding of spatially-resolved biological mechanisms.

Conclusions:

• SORBET effectively maps spatial and marker information from spatial omics technologies to emergent biological phenotypes.
• The developed framework provides novel techniques for identifying biological processes contributing to predicted phenotypes.
• This approach enhances the analysis of complex spatial omics data for biological discovery.