File size: 22,720 Bytes
362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 8f91d7b 362d677 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 | {
"@context": {
"@language": "en",
"@vocab": "https://schema.org/",
"sc": "https://schema.org/",
"cr": "http://mlcommons.org/croissant/",
"rai": "http://mlcommons.org/croissant/RAI/",
"dct": "http://purl.org/dc/terms/",
"annotation": "cr:annotation",
"citeAs": "cr:citeAs",
"conformsTo": "dct:conformsTo",
"containedIn": "cr:containedIn",
"data": {"@id": "cr:data", "@type": "@json"},
"dataType": {"@id": "cr:dataType", "@type": "@vocab"},
"field": "cr:field",
"fileObject": "cr:fileObject",
"fileSet": "cr:fileSet",
"includes": "cr:includes",
"recordSet": "cr:recordSet",
"source": "cr:source",
"subField": "cr:subField",
"prov": "http://www.w3.org/ns/prov#"
},
"@type": "sc:Dataset",
"name": "scShapeBench",
"description": "A benchmark dataset for single-cell shape analysis with four configurations: (1) scRNAseq — 2,547,517 cells across 102 real-world scRNA-seq datasets in AnnData (.h5ad) format with precomputed PCA embeddings and Leiden clustering; (2) synthetic — synthetically generated single-cell data in NumPy compressed (.npz) format with per-sample JSON metadata; (3) annotations — per-dataset multi-label shape annotations from 9 independent expert annotators; (4) labels — aggregated shape labels derived via majority vote, soft aggregation, confidence weighting, and union across the 9 annotators.",
"url": "https://huggingface.co/datasets/scShape-Benchmark/scShapeBench",
"license": "https://creativecommons.org/licenses/by/4.0/",
"conformsTo": "http://mlcommons.org/croissant/1.1",
"datePublished": "2026-05-05",
"version": "1.0.0",
"citeAs": "",
"keywords": ["single-cell", "RNA-seq", "scRNA-seq", "h5ad", "AnnData", "genomics", "benchmark", "shape-analysis", "synthetic"],
"rai:hasSyntheticData": true,
"prov:wasDerivedFrom": [
{ "@id": "https://cellxgene.cziscience.com" },
{ "@id": "https://www.10xgenomics.com/datasets" },
{ "@id": "https://singlecell.broadinstitute.org/single_cell" },
{ "@id": "https://www.ebi.ac.uk/gxa/sc" }
],
"prov:wasGeneratedBy": [
{
"@type": "prov:Activity",
"name": "Data Collection",
"prov:used": [
{ "@id": "https://cellxgene.cziscience.com" },
{ "@id": "https://www.10xgenomics.com/datasets" },
{ "@id": "https://singlecell.broadinstitute.org/single_cell" },
{ "@id": "https://www.ebi.ac.uk/gxa/sc" }
],
"description": "102 real-world scRNA-seq datasets were manually curated from four public repositories: CELLxGENE Census (62 datasets), 10x Genomics public dataset catalog (30 datasets), Broad Institute Single Cell Portal (8 datasets), and EMBL-EBI Single Cell Expression Atlas (1 dataset). Datasets were selected to cover all four structural shape classes (Clusters, Single-Trajectory, Multi-Branching, Archetypal). Per-dataset source URLs, accession identifiers, and associated publication references are documented in the accompanying datasheet."
},
{
"@type": "prov:Activity",
"name": "Synthetic Data Generation",
"description": "Synthetic single-cell data were generated using a dedicated software tool described in the accompanying paper. The generation process, parameters, and seeds are fully documented therein. The source code repository will be made publicly available post-publication."
},
{
"@type": "prov:Activity",
"name": "Data Preprocessing",
"description": "All scRNAseq datasets were processed through a uniform Scanpy pipeline: (1) quality-control filtering, (2) log-normalization (scanpy.pp.normalize_total followed by scanpy.pp.log1p), (3) subsampling to a maximum of 50,000 cells per dataset, (4) PCA dimensionality reduction, and (5) Leiden graph-based clustering (resolution=1.0). 2D PHATE and UMAP embeddings were computed exclusively for expert annotation and are not included in the benchmarked inputs. Raw integer counts are preserved in the .raw slot of each h5ad file."
},
{
"@type": "prov:Activity",
"name": "Data Annotation",
"description": "Nine expert biologists and bioinformaticians independently annotated each dataset via a dedicated web gallery displaying 2D PHATE and UMAP projections. Annotators assigned multi-label structural categories from the taxonomy {Clusters, Single-Trajectory, Multi-Branching, Archetypal}. No communication between annotators was permitted during annotation. Aggregated labels were derived using four strategies: majority vote (label present if ≥5 of 9 annotators selected it), soft aggregation (fraction of annotators), confidence-weighted aggregation, and union. Inter-annotator agreement: mean pairwise Jaccard similarity = 0.403; per-label Fleiss' kappa range = 0.161–0.255."
}
],
"rai:dataCollection": "The scRNAseq configuration comprises real-world single-cell RNA sequencing datasets manually curated from four public repositories: CELLxGENE Census hosted by the Chan Zuckerberg Initiative (62 datasets), the 10x Genomics public dataset catalog (30 datasets), the Broad Institute Single Cell Portal (8 datasets), and the EMBL-EBI Single Cell Expression Atlas (1 dataset). The synthetic configuration contains computationally generated single-cell data providing ground-truth topology for controlled benchmarking. Per-dataset source URLs and associated publication references are provided in the accompanying datasheet.",
"rai:dataCollectionType": ["Existing Datasets", "Synthetic"],
"rai:dataCollectionRawData": "Raw integer counts are preserved in the .raw slot of each h5ad file alongside the preprocessed log-normalized counts in .X. The source repository and original publication are recorded per-dataset in the accompanying datasheet.",
"rai:dataPreprocessingProtocol": "All scRNAseq datasets were processed through a uniform Scanpy pipeline comprising quality-control filtering, log-normalization, and subsampling to a fixed 50,000 cells per dataset. PCA embeddings and Leiden clustering were computed at uniform resolution across all datasets. 2D PHATE and UMAP embeddings were computed solely for expert annotation and are not included in the benchmarked inputs.",
"rai:dataAnnotationProtocol": "Nine expert biologists and bioinformaticians independently assigned multi-label structural annotations from the taxonomy {Clusters, Single-Trajectory, Multi-Branching, Archetypal} by visually inspecting 2D PHATE and UMAP embeddings via a dedicated web gallery. A structural class is assigned to a dataset if at least 5 of 9 annotators selected it (majority-support rule). Datasets with no majority-supported class are excluded from the primary benchmark, yielding 79 high-support datasets out of 102.",
"rai:annotationsPerItem": "9",
"rai:dataAnnotationAnalysis": "Inter-annotator agreement was assessed at both label-set and per-label levels. Mean pairwise Jaccard similarity was 0.403. Per-label Fleiss' kappa ranged from 0.161 to 0.255, reflecting the inherent ambiguity of assigning global shape categories to real scRNAseq data where multiple structural regimes may coexist simultaneously.",
"rai:personalSensitiveInformation": "The dataset contains gene expression profiles derived from human and non-human cells. All data originate from previously published public repositories; no individual-level identifiers are included. All source datasets were released by their original authors under open-access terms permitting redistribution. Per-dataset source licenses are recorded in the accompanying datasheet.",
"rai:dataBiases": "Dataset selection was guided by coverage of four structural shape classes, introducing sampling bias toward datasets judged likely to exhibit geometric structure. The 10x Genomics catalog is weighted toward PBMC datasets that tend to produce cluster-like geometry, while CELLxGENE and the Broad SCP contribute differentiation, disease, and tissue-atlas studies. Expert annotations were performed on 2D PHATE and UMAP projections, which may not faithfully represent high-dimensional manifold structure and may introduce visual bias. Moderate inter-annotator agreement (mean Jaccard 0.403) indicates residual subjectivity in label assignment.",
"rai:dataLimitations": "All scRNAseq datasets are subsampled to 50,000 cells, which may alter the geometric structure of very large datasets. Expert structural labels reflect visual consensus on 2D embeddings and do not constitute mathematical ground-truth topology. The synthetic configuration provides controlled ground truth but may not capture the full complexity of real biological data.",
"rai:dataUseCases": "scShapeBench is intended for benchmarking computational methods that recover the global geometric or topological structure of single-cell gene expression data. It should not be used for clinical diagnosis, treatment decisions, or any application requiring individual-level genomic inference.",
"rai:dataSocialImpact": "This benchmark advances the development and fair comparison of computational methods for single-cell biology, contributing to basic research in genomics and cell biology. The dataset does not contain identifiable personal information and is not intended for surveillance, clinical, or forensic applications.",
"rai:dataMaintenancePlan": "Issues, errors, or questions regarding the data can be reported via the Hugging Face repository discussion tab. The dataset will be updated to include additional configurations and corrected annotations as the associated research progresses.",
"distribution": [
{
"@type": "cr:FileSet",
"@id": "h5ad-files",
"name": "h5ad-files",
"description": "AnnData files containing single-cell gene expression data with PCA embeddings and Leiden clustering",
"includes": "data/scRNAseq/SCD-*.h5ad",
"encodingFormat": "application/x-hdf5"
},
{
"@type": "cr:FileSet",
"@id": "npz-files",
"name": "npz-files",
"description": "NumPy compressed arrays containing synthetic single-cell data",
"includes": "data/synthetic/sample_*.npz",
"encodingFormat": "application/zip"
},
{
"@type": "cr:FileSet",
"@id": "json-metadata-files",
"name": "json-metadata-files",
"description": "Per-sample JSON metadata files for synthetic data",
"includes": "data/synthetic/sample_*.json",
"encodingFormat": "application/json"
},
{
"@type": "cr:FileObject",
"@id": "cell-metadata-csv",
"name": "cell_metadata.csv",
"description": "Combined cell-level metadata from all datasets: number of genes detected, Leiden cluster assignment, and source file identifier",
"contentUrl": "cell_metadata.csv",
"encodingFormat": "text/csv",
"sha256": "c8d0a6827425a212172cbef11866aa9e598c154c3a513a3f46f1fa3577671d94"
},
{
"@type": "cr:FileObject",
"@id": "gene-metadata-csv",
"name": "gene_metadata.csv",
"description": "Combined gene-level metadata from all datasets: gene IDs, feature types, dispersion statistics, and highly-variable flags",
"contentUrl": "gene_metadata.csv",
"encodingFormat": "text/csv",
"sha256": "decc2d36c8a1819de6b64ae847eabda50c8dfa60afd28c23d7fd842a4746bbab"
},
{
"@type": "cr:FileObject",
"@id": "dataset-index-csv",
"name": "dataset_index.csv",
"description": "Per-file summary statistics: number of cells, number of genes, and file size",
"contentUrl": "dataset_index.csv",
"encodingFormat": "text/csv",
"sha256": "fbcd479d9cd12d6898d3cae4eff5935d7aaacb013cc52cf19c91d4565cabf5dd"
},
{
"@type": "cr:FileObject",
"@id": "annotations-parquet",
"name": "annotations.parquet",
"description": "Per-dataset multi-label shape annotations from 9 independent expert annotators, one row per dataset-annotator pair",
"contentUrl": "labels/annotations.parquet",
"encodingFormat": "application/vnd.apache.parquet",
"sha256": "af00c1db034639c3dbc4530f6f9ae12ef270ba5974d56a55c1bda6e9c3746992"
},
{
"@type": "cr:FileObject",
"@id": "labels-parquet",
"name": "labels.parquet",
"description": "Aggregated shape labels from 4 strategies (majority, soft, confidence_weighted, union), one row per dataset-aggregation pair",
"contentUrl": "labels/labels.parquet",
"encodingFormat": "application/vnd.apache.parquet",
"sha256": "8c486a6db38128871911c4e6c2d262411648240b07edd00eec99c0d96582fc08"
}
],
"recordSet": [
{
"@type": "cr:RecordSet",
"name": "dataset_index",
"description": "Summary statistics for each of the 105 single-cell datasets in the benchmark",
"field": [
{
"@type": "cr:Field",
"name": "file_id",
"description": "Dataset identifier (e.g., SCD-0001)",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "dataset-index-csv"},
"extract": {"column": "file_id"}
}
},
{
"@type": "cr:Field",
"name": "filename",
"description": "Filename of the h5ad file",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "dataset-index-csv"},
"extract": {"column": "filename"}
}
},
{
"@type": "cr:Field",
"name": "n_cells",
"description": "Number of cells in the dataset",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "dataset-index-csv"},
"extract": {"column": "n_cells"}
}
},
{
"@type": "cr:Field",
"name": "n_genes",
"description": "Number of genes in the dataset",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "dataset-index-csv"},
"extract": {"column": "n_genes"}
}
},
{
"@type": "cr:Field",
"name": "file_size_bytes",
"description": "File size in bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "dataset-index-csv"},
"extract": {"column": "file_size_bytes"}
}
}
]
},
{
"@type": "cr:RecordSet",
"name": "cell_metadata",
"description": "Cell-level metadata across all 2,547,517 cells in the benchmark",
"field": [
{
"@type": "cr:Field",
"name": "cell_barcode",
"description": "Unique cell barcode identifier",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "cell-metadata-csv"},
"extract": {"column": "cell_barcode"}
}
},
{
"@type": "cr:Field",
"name": "n_genes",
"description": "Number of genes detected in the cell",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "cell-metadata-csv"},
"extract": {"column": "n_genes"}
}
},
{
"@type": "cr:Field",
"name": "file_id",
"description": "Source dataset identifier (foreign key to dataset_index.file_id)",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "cell-metadata-csv"},
"extract": {"column": "file_id"}
},
"references": {
"field": {"@id": "dataset_index/file_id"}
}
}
]
},
{
"@type": "cr:RecordSet",
"name": "gene_metadata",
"description": "Gene-level metadata across all datasets including IDs, dispersion statistics, and highly-variable flags",
"field": [
{
"@type": "cr:Field",
"name": "gene_index",
"description": "Gene identifier (Ensembl ID or symbol)",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "gene-metadata-csv"},
"extract": {"column": "gene_index"}
}
},
{
"@type": "cr:Field",
"name": "gene_ids",
"description": "Ensembl gene identifier",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "gene-metadata-csv"},
"extract": {"column": "gene_ids"}
}
},
{
"@type": "cr:Field",
"name": "highly_variable",
"description": "Whether the gene is flagged as highly variable",
"dataType": "sc:Boolean",
"source": {
"fileObject": {"@id": "gene-metadata-csv"},
"extract": {"column": "highly_variable"}
}
},
{
"@type": "cr:Field",
"name": "means",
"description": "Mean expression level",
"dataType": "sc:Float",
"source": {
"fileObject": {"@id": "gene-metadata-csv"},
"extract": {"column": "means"}
}
},
{
"@type": "cr:Field",
"name": "n_cells",
"description": "Number of cells expressing this gene",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "gene-metadata-csv"},
"extract": {"column": "n_cells"}
}
},
{
"@type": "cr:Field",
"name": "file_id",
"description": "Source dataset identifier (foreign key to dataset_index.file_id)",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "gene-metadata-csv"},
"extract": {"column": "file_id"}
},
"references": {
"field": {"@id": "dataset_index/file_id"}
}
}
]
},
{
"@type": "cr:RecordSet",
"name": "annotations",
"description": "Per-dataset multi-label shape annotations from 9 independent expert annotators. One row per dataset-annotator pair; binary columns indicate which shape categories were assigned.",
"field": [
{
"@type": "cr:Field",
"name": "dataset_id",
"description": "Dataset identifier (e.g., SCD-0001)",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "annotations-parquet"},
"extract": {"column": "dataset_id"}
},
"references": {
"field": {"@id": "dataset_index/file_id"}
}
},
{
"@type": "cr:Field",
"name": "annotator_id",
"description": "Annotator index (0-8)",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "annotations-parquet"},
"extract": {"column": "annotator_id"}
}
},
{
"@type": "cr:Field",
"name": "archetypal",
"description": "1 if the annotator assigned the archetypal shape category, 0 otherwise",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "annotations-parquet"},
"extract": {"column": "archetypal"}
}
},
{
"@type": "cr:Field",
"name": "multi_branch",
"description": "1 if the annotator assigned the multi_branch shape category, 0 otherwise",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "annotations-parquet"},
"extract": {"column": "multi_branch"}
}
},
{
"@type": "cr:Field",
"name": "simple_traj",
"description": "1 if the annotator assigned the simple_traj shape category, 0 otherwise",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "annotations-parquet"},
"extract": {"column": "simple_traj"}
}
},
{
"@type": "cr:Field",
"name": "clusters",
"description": "1 if the annotator assigned the clusters shape category, 0 otherwise",
"dataType": "sc:Integer",
"source": {
"fileObject": {"@id": "annotations-parquet"},
"extract": {"column": "clusters"}
}
}
]
},
{
"@type": "cr:RecordSet",
"name": "labels",
"description": "Aggregated shape labels derived from 9 annotators using four strategies. One row per dataset-aggregation pair; values are floats in [0, 1] representing label confidence per shape category.",
"field": [
{
"@type": "cr:Field",
"name": "dataset_id",
"description": "Dataset identifier (e.g., SCD-0001)",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "labels-parquet"},
"extract": {"column": "dataset_id"}
},
"references": {
"field": {"@id": "dataset_index/file_id"}
}
},
{
"@type": "cr:Field",
"name": "aggregation",
"description": "Aggregation strategy: majority, soft, confidence_weighted, or union",
"dataType": "sc:Text",
"source": {
"fileObject": {"@id": "labels-parquet"},
"extract": {"column": "aggregation"}
}
},
{
"@type": "cr:Field",
"name": "archetypal",
"description": "Aggregated label score for the archetypal shape category",
"dataType": "sc:Float",
"source": {
"fileObject": {"@id": "labels-parquet"},
"extract": {"column": "archetypal"}
}
},
{
"@type": "cr:Field",
"name": "multi_branch",
"description": "Aggregated label score for the multi_branch shape category",
"dataType": "sc:Float",
"source": {
"fileObject": {"@id": "labels-parquet"},
"extract": {"column": "multi_branch"}
}
},
{
"@type": "cr:Field",
"name": "simple_traj",
"description": "Aggregated label score for the simple_traj shape category",
"dataType": "sc:Float",
"source": {
"fileObject": {"@id": "labels-parquet"},
"extract": {"column": "simple_traj"}
}
},
{
"@type": "cr:Field",
"name": "clusters",
"description": "Aggregated label score for the clusters shape category",
"dataType": "sc:Float",
"source": {
"fileObject": {"@id": "labels-parquet"},
"extract": {"column": "clusters"}
}
}
]
}
]
}
|