STANCE: a unified statistical model to detect cell-type-specific spatially variable genes in spatial transcriptomics

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Detecting spatially variable genes (SVGs) is crucial in spatial transcriptomics.
• Existing methods for cell type-specific SVGs (ctSVGs) can yield rotation-dependent results.
• The relationship between SVGs and ctSVGs can be complex and variable.

Purpose of the Study:

• To develop a unified statistical model for detecting both SVGs and ctSVGs.
• To address the limitations of existing methods, particularly their dependence on tissue orientation.
• To provide robust and rotation-invariant results in spatial transcriptomics analysis.

Main Methods:

• Proposed STANCE, a unified statistical model using a linear mixed-effect framework.
• Integrated gene expression, spatial location, and cell type composition data.
• Employed a two-stage approach: initial SVG/ctSVG detection and subsequent ctSVG-specific testing.

Main Results:

• STANCE demonstrated robust performance in simulations and real data analyses.
• The model ensures tissue rotation-invariant results for SVG and ctSVG detection.
• Validated through extensive simulations and public spatial transcriptomics datasets.

Conclusions:

• STANCE offers a novel and improved approach for identifying SVGs and ctSVGs.
• The method enhances the reliability and interpretability of spatial transcriptomics data.
• STANCE has significant potential for advancing downstream analyses in the field.