Spatially Variable Genes Benchmark

OmniBenchmark implementation of the OpenProblems spatially variable genes task. Evaluates 16 SVG detection methods on 50 spatial transcriptomics datasets using Kendall rank correlation.

Methods

BOOST-GP, GPcounts, Moran's I, nnSVG, scGCO, SEPAL, SomDE, SpaGCN, SpaGFT, SpaNVE, SPARK, SPARK-X, SpatialDE, SpatialDE2, random_ranking, true_ranking

Data

50 preprocessed spatial transcriptomics datasets (H5AD format) available from OpenProblems. See datasets documentation for download instructions.

Usage

Run the full benchmark:

ob run benchmark -b spatially_variable_genes_apptainer.yml --local-storage

Preview execution without running:

ob run benchmark -b spatially_variable_genes_apptainer.yml --dry

Run in parallel with 8 cores:

ob run benchmark -b spatially_variable_genes_apptainer.yml --local-storage --cores 8

Requirements

• OmniBenchmark (0.3+)
• Apptainer/Singularity (v1.0+)
• 70 GB disk space

Related Repositories

• Metric Collector - Aggregates scores from all methods and datasets

References

• OpenProblems SVG Task
• OmniBenchmark