BatchSVG

BatchSVG identifies batch-biased spatially variable genes (SVGs) in spatial transcriptomics data by comparing per-gene deviance ranks under a binomial model with and without a batch covariate.

Key Features:

• Batch-biased SVG detection: Identifies genes whose spatial variability is associated with known biases or technical artifacts such as slide, capture area, sample, or donor sex.
• Rank comparison approach: Compares per-gene deviance ranks under a binomial model fit with and without inclusion of a batch-associated covariate to flag significantly rank-shifted genes.
• Flexible batch variable definition: Supports user-defined batch variables including sample, donor sex, or other batch effects of interest.

Scientific Applications:

• Spatial atlas generation: Filters batch-biased SVGs from large-scale spatial transcriptomics atlases spanning hundreds of donors to obtain a common set of SVGs for downstream analysis.
• Spatial domain detection improvement: Removes technically confounded SVGs to improve performance in downstream analyses such as spatial domain detection.

Methodology:

BatchSVG is based on the binomial deviance model (Townes et al, 2019). For each gene, the method fits a binomial model (i) with and (ii) without a covariate associated with the known bias or technical artifact, then compares the per-gene deviance ranks between the two model fits to identify significantly rank-shifted genes.