bjames/striatum-snrna-qc

Human Striatum snRNA-seq Replicate Detection Dataset

109 single-nucleus RNA-seq (snRNA-seq) samples from the human striatum, used in the replicate-detection-snrna benchmark task from Terminal Bench Science.

Contents

.
├── input/                              # 109 H5AD files (one per sample)
│   ├── 190726_HD_snRNA-H4.h5ad
│   ├── 190726_HD_snRNA-H6.h5ad
│   └── ... (109 total)
└── verification/
    └── sample_metadata_s2.tsv          # Sample metadata derived from Cell paper Supplementary Data S2

input/

109 H5AD files, each containing a single snRNA-seq sample from the human striatum. Files are flat (no subdirectory structure). Sample names encode batch/date information but not donor identity (by design, donor detection is the task).

Each H5AD contains:

• adata.X: sparse (CSR) gene-by-cell expression matrix (integer counts)
• adata.var: gene annotations
• adata.obs: per-cell metadata

verification/sample_metadata_s2.tsv

A 13-column TSV (109 rows) extracted from the Cell paper's Supplementary Data S2 "Sample metadata" sheet. This is a full extract, so reviewers can independently verify provenance against the published supplementary table.

Column	Description
Sample	Sample name (matches H5AD filename minus .h5ad)
donor_id	Donor identifier
region	Brain region: NAc or Putamen
Age	Donor age at death (years)
Sex	Donor sex: M or F
PMI	Post-mortem interval (hours)
source	Tissue source
prep	Library preparation protocol (e.g. 10x v3, 10x multiome)
input_type	Library input type: bulk or BVE (blood-vessel-enriched)
ZPMI	Z-scored post-mortem interval
annotated neurons (#)	Number of annotated neurons in the sample
MSN (%)	Fraction of medium spiny neurons
Excitatory (%)	Fraction of excitatory neurons

The donor_id and input_type columns provide an external cross-reference for the benchmark's hidden donor ground truth. The verifier image's build-time cross-check asserts that every sample's donor_id and input_type in donor_truth.tsv matches this S2 extract exactly. The remaining 10 columns (region, Age, Sex, PMI, etc.) are not used by the cross-check but are included for reviewer transparency and independent verifiability: for example, Sex can be confirmed from Y-chromosome gene expression (UTY/KDM5D/DDX3Y) in the H5AD data.

Citation

This dataset will be described in an upcoming Cell publication (expected online 1 September 2026):

Linville*, R. M., James*, B. T., Galani, K., Ho, L., Shin, J., Oliver, E., Bock, R., Murray, E. M., Louca, M., Oke, O., Wang, C., Engelberg-Cook, E., DeTure, M., Farrell, V., Fitzwalter, B., Pineda, S., Sathitloetsakun, S., Liang, X., Madras, B., Dickson, D. W., Mash, D. C., Turecki, G., Wheeler, V. C., Alvarez, V. A., Gabuzda, D., Kellis, M., Heiman, M. (2026). "Cross-species single cell atlas of the striatum defines cell type and subregion disease vulnerabilities." Cell (2026, in press).

Until the paper is published, please cite as:

James, B. (MIT). Human striatum single-nucleus RNA-seq dataset. 2026. Hugging Face dataset: bjames/replicate-detection-snrna.

License

This dataset is released under CC BY 4.0.

Data use note

Pre-publication use of this data for benchmark evaluation is encouraged. The accompanying Cell paper (expected online 1 September 2026) will provide full methodological context.