Addressing the mean-variance relationship in spatially resolved transcriptomics data with spoon

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatially resolved transcriptomics (SRT) enables gene expression analysis within tissue context.
• Identifying spatially variable genes (SVGs) is key for understanding tissue organization and function.
• Existing methods for SVG ranking may be affected by technical biases, particularly log-transformation artifacts impacting the mean-variance relationship.

Purpose of the Study:

• To address the technical bias in log-transformed SRT data that affects the mean-variance relationship.
• To develop a robust statistical framework for accurate identification and prioritization of SVGs.
• To introduce a novel method, spoon, for bias correction in SRT data analysis.

Main Methods:

• Demonstration of the mean-variance relationship in SRT data.
• Development of the spoon statistical framework utilizing empirical Bayes techniques.
• Application and validation of spoon on both simulated and real SRT datasets.

Main Results:

• Confirmation of the mean-variance relationship bias in SRT data.
• spoon effectively removes the identified bias, leading to more accurate SVG prioritization.
• spoon demonstrates superior performance compared to existing methods in simulated and real data.

Conclusions:

• The proposed spoon framework provides a more accurate approach to identifying SVGs in SRT data.
• spoon's empirical Bayes method corrects for log-transformation bias, enhancing biological insights from spatial transcriptomics.
• A publicly available software implementation of spoon facilitates its adoption in the research community.