SpatialSort: a Bayesian model for clustering and cell population annotation of spatial proteomics data

Area of Science:

• Single-cell spatial proteomics
• Computational biology
• Bioinformatics

Background:

• Spatial proteomics technologies enable in situ profiling of proteins in thousands of single cells.
• Current clustering methods often ignore spatial context and prior biological knowledge, limiting tissue analysis.
• Understanding spatial relationships between cells is crucial for advancing tissue composition studies.

Purpose of the Study:

• To develop a novel method, SpatialSort, that incorporates spatial context and prior biological knowledge for improved clustering and annotation of spatial proteomics data.
• To address the limitations of existing clustering approaches that focus solely on expression values.
• To enable automated annotation of cell populations within complex tissue microenvironments.

Main Methods:

• Developed SpatialSort, a spatially aware Bayesian clustering approach.
• Incorporated prior biological knowledge about expected cell populations.
• Accounted for cell-type specific spatial neighbor affinities.
• Validated using synthetic and real spatial proteomics datasets.

Main Results:

• SpatialSort significantly improves clustering accuracy by leveraging spatial and prior information.
• The method enables automated annotation of cell clusters.
• Demonstrated successful label transfer between spatial and non-spatial data modalities.
• Applied to a diffuse large B-cell lymphoma dataset, showcasing real-world utility.

Conclusions:

• SpatialSort offers a powerful approach for analyzing spatial proteomics data.
• Integrating spatial context and prior knowledge enhances the understanding of tissue architecture.
• The method facilitates automated cell population identification and annotation in complex biological systems.