BatchSVG: identifying batch-biased genes in the application of spatially variable gene detection

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 crucial for understanding tissue organization.
• Current SVG detection methods often analyze tissue sections independently, limiting large-scale atlas analysis.

Purpose of the Study:

• To introduce BatchSVG, a novel tool for identifying and removing batch-biased genes in SRT data.
• To address the challenge of technical artifacts (e.g., slide, capture area) that can confound SVG identification.
• To enhance the performance of downstream analyses, such as spatial domain detection, by improving SVG selection.

Main Methods:

• BatchSVG compares per-gene deviance ranks from binomial models with and without batch-effect covariates.
• Genes with significant rank changes between models are flagged as batch-biased.
• The method was evaluated on two SRT datasets to assess its effectiveness.

Main Results:

• BatchSVG successfully identifies genes associated with known technical biases in SRT data.
• Removal of batch-biased genes led to improved results in downstream spatial domain detection.
• The tool offers a robust approach to mitigate technical artifacts in large-scale spatial atlases.

Conclusions:

• BatchSVG is an effective tool for identifying and removing batch-biased genes in spatially resolved transcriptomics.
• Mitigating technical artifacts through BatchSVG enhances the reliability of spatially variable gene identification.
• This approach facilitates more accurate downstream analyses and the construction of robust spatial atlases.