Identifying spatially variable genes by projecting to morphologically relevant curves.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics provides high-resolution gene expression data with spatial context.
• Identifying spatially variable genes is crucial for understanding tissue organization.
• Current 2D methods struggle with implicitly 1D tissue structures, overlooking underlying morphology.

Purpose of the Study:

• To develop a new methodology for spatial transcriptomics data analysis that accounts for underlying tissue morphology.
• To improve the identification of spatially variable genes in complex tissue structures.
• To establish a more accurate coordinate system for analyzing gene expression patterns.

Main Methods:

• Utilized spectral graph theory to derive a 1D curve approximating sample coordinates.
• Established a new, morphologically relevant coordinate system.
• Developed a generalized additive model (GAM) to detect gene expression variability in the new coordinate system.
• Directly modeled gene counts, avoiding normalization and transformation steps.

Main Results:

• Demonstrated improved performance compared to existing hypothesis testing methods.
• Accurately estimated gene expression patterns.
• Precisely identified spatial loci with significant gene expression deviations.
• Validated the approach through simulations and multi-platform experimental data (Slide-seq, MERFISH).

Conclusions:

• The proposed methodology effectively captures 1D tissue organization within spatial transcriptomics data.
• This morphologically relevant coordinate system enhances the detection of spatially variable genes.
• The approach offers a robust and accurate alternative for analyzing spatial gene expression patterns across different platforms.