Dataset Details
Dataset Description
SOHIB-Corpus is a curated and manually annotated benchmark dataset for evaluating spatial horizontal integration methods. The corpus spans over 160 tissue sections and more than 3.5 million cells and spots, drawn from four spatial omics modalities (transcriptomics, proteomics, metabolomics, and epigenomics) and 12 measurement platforms, organized into 33 integration tasks. It was built to support the spatial omics horizontal integration benchmarking toolkit SOHIB.
The corpus is organized into 33 integration tasks spanning:
- Sequencing-based spatial transcriptomics (sST): 10x Visium, Stereo-seq, Array-seq, Slide-seq
- Imaging-based spatial transcriptomics (iST): CosMx, MERFISH, STARmap, 10x Xenium
- Spatial proteomics: CODEX
- Spatial metabolomics: MALDI-MSI
- Spatial epigenomics: MERFISH-epigenomic (H3K27ac, H3K4me3), MISAR-seq (ATAC)
Sections were collected from human, mouse, and axolotl specimens across eight tissue types.
A dedicated cross-platform mouse brain reference is also included: five coronal sections profiled by MERFISH, STARmap, and Slide-seq, annotated with unified Allen Common Coordinate Framework (CCF) region labels, enabling evaluation across a controlled gradient of cross-platform batch effect.
Each section is annotated with:
- Spatial domain labels, collected from published annotations or assigned through manual annotation (
adata.obs["SOHIB_domain_key"]) - Cell-type labels, for datasets acquired at single-cell resolution, as well as for Slide-seq datasets, whose spot diameter approaches single-cell resolution (
adata.obs["SOHIB_cell_type_key"])
Related Sources
- Repository: https://github.com/hannshu/SOHIB
Acknowledgement
We thank the original authors and research groups who generated and publicly released the datasets incorporated into SOHIB-Corpus.
- Downloads last month
- 5