Dimensionality reduction for visualizing spatially resolved profiling data using SpaSNE

Area of Science:

• Single-cell biology
• Genomics
• Bioinformatics

Background:

• Spatially resolved profiling technologies offer comprehensive molecular characterization.
• Dimensionality reduction is crucial for analyzing spatially resolved data.
• Existing methods like t-SNE and UMAP are not optimized for spatial data.

Purpose of the Study:

• To develop a dimensionality reduction method tailored for spatially resolved profiling data.
• To integrate both spatial and molecular information for enhanced data analysis.
• To improve the visualization and interpretation of complex biological datasets.

Main Methods:

• Developed a novel spatially resolved t-SNE (SpaSNE) method.
• Applied SpaSNE to diverse public datasets from multiple experimental platforms (Visium, STARmap, MERFISH).
• Compared SpaSNE performance against t-SNE and UMAP using diseased and normal tissue data.

Main Results:

• SpaSNE effectively integrates spatial and molecular information.
• SpaSNE provides more accurate and meaningful visualizations compared to t-SNE and UMAP.
• The method successfully elucidates underlying spatial and molecular data structures across various tissues and cell types.

Conclusions:

• SpaSNE enables reliable interpretation of cell types using combined molecular and spatial data.
• This method provides a foundation for downstream analyses like differential gene expression and trajectory analysis.
• SpaSNE demonstrates broad applicability for robust analysis of spatially resolved profiling data.