{"success":true,"data":{"author":"Lin Zhong, Carsen Stringer, Marius Pachitariu et al. (HHMI Janelia Research Campus); subset by Kristin Branson","bucketAccessType":"public","bucketName":"kristinbranson/neurodata-reuse-zhong2025","bucketPrefix":"","category":"Neuroscience","cloudProvider":"huggingface","createdAt":"2026-11-22T01:09:18.589Z","description":"A size-reduced copy of the two-photon calcium imaging dataset from Zhong, Baptista, Gattoni, Arnold, Flickinger, Stringer and Pachitariu, Unsupervised pretraining in biological neural networks (Nature 2025, Janelia Research Campus), which recorded very large neural populations in mouse visual cortex to study how unsupervised exposure shapes learning. All 19 mice, 89 sessions and 38,110 trials of the 412 GB Janelia release (DOI 10.25378/janelia.28811129.v2) are kept, but each session keeps a deterministic random 9.8% of cells, 459,723 of 4.69 million neurons, bringing it to about 47 GB. spk/ holds deconvolved neural activity per session (89 files, 40 GB), beh/ the unmodified behavioral data (28 files, 6.6 GB), and retinotopy/ retinotopic maps aligned to the kept cells; DATALIMIT_SUBSET.csv records kept versus total cells per session. Per-session cell counts are about a tenth of those in the paper. Load the NumPy files with numpy. Uses: large-scale population coding, representational drift and learning, and neural data foundation models. Packaged by Kristin Branson for a neural-data reuse benchmark; the original and this copy are CC BY 4.0. Cite Zhong et al., Nature 2025 (doi 10.1038/s41586-025-09180-y) and the Janelia dataset.","displayTitle":"Unsupervised Pretraining in Biological Neural Networks: Mouse Visual Cortex Two-Photon Calcium Imaging of 19 Mice, Neuron-Subsampled Copy (Zhong, Pachitariu, Janelia)","huggingFaceUrl":"https://huggingface.co/datasets/kristinbranson/neurodata-reuse-zhong2025","id":"67f694ed-3c7f-5e47-b199-068c23574afb","license":"CC-BY-4.0","licenseEvidenceUrl":"https://api.datacite.org/dois/10.25378/janelia.28811129.v2","objectCountEstimate":209,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1038/s41586-025-09180-y","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"kristinbranson/neurodata-reuse-zhong2025\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"kristinbranson/neurodata-reuse-zhong2025\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=kristinbranson, license:cc-by-4.0, size_categories:10B<n<100B, downloads=584"],"format":"mixed","generatedAt":"2026-11-22T01:10:18.318Z","layout":{"example":"DATALIMIT_SUBSET.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"session,n_cells_total,n_cells_kept\nDR10_2022_07_12_1,58224,5706\nDR10_2022_07_19_1,54741,5365\nDR10_2022_07_21_1,50347,4934\nDR10_2022_07_28_1,50195,4919\nDR10_2022_07_29_1,56106,5498\nDR10_2022_07_30_1,47963,4700\nDR15_2022_10_09_1,55746,5463\nDR15_2022_10_19_1,58930,5775\nDR15_2022_10_20_1,64130,6285\nDR15_2022_11_02_1,57540,5639\nDR15_2022_11_03_2,59251,5807\nLZ13_2024_05_15_1,34658,3396\nLZ13_2024_05_16_2,34575,3388\nLZ13_2024_05_28_1,38489,3772\nLZ13_2024_05_29_1,39358,3857\nLZ16_2024_05_18_2,33577,3291\nLZ16_2024_05_","encoding":"utf8","key":"DATALIMIT_SUBSET.csv","truncatedAt":512}],"sampleObjects":["DATALIMIT_SUBSET.csv","beh/Beh_naive_test1.npy","beh/Beh_naive_test2.npy","beh/Beh_naive_test3.npy","beh/Beh_sup_test1.npy"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\ntask_categories:\n  - other\ntags:\n  - neuroscience\n  - two-photon calcium imaging\n  - mouse\n  - visual cortex\n  - benchmark\npretty_name: Zhong 2025 unsupervised pretraining (neuron-subsampled)\nsize_categories:\n  - 10B<n<100B\n---\n\n# Zhong et al. 2025, unsupervised pretraining — neuron-subsampled subset\n\nA size-reduced copy of the two-photon calcium imaging dataset released with\nZhong et al., *Unsupervised pretraining in biological neural networks*\n(Nature, 2025). It exists to make the dataset usable as a benchmark task under a\n50 GB cap. **It is not a new dataset** — every value here comes unmodified from\nthe published release.\n\nIf you use this data, **cite the original paper and dataset**, not this copy.\n\n## Citation\n\n> Zhong, L., Baptista, S., Gattoni, R., Arnold, J., Flickinger, D., Stringer, C.,\n> & Pachitariu, M. (2025). Unsupervised pretraining in biological neural networks.\n> *Nature*. https://doi.org/10.1038/s41586-025-09180-y\n\nDataset:\n\n> Zhong, L., Baptista, S., Gattoni, R., Arnold, J., Flickinger, D., Stringer, C.,\n> & Pachitariu, M. (2025). Unsupervised pretraining in biological neural network\n> [Dataset]. Janelia Research Campus.\n> https://doi.org/10.25378/janelia.28811129.v2\n\n```bibtex\n@article{zhong2025unsupervised,\n  title   = {Unsupervised pretraining in biological neural networks},\n  author  = {Zhong, Lin and Baptista, Scott and Gattoni, Rachel and Arnold, Jon\n             and Flickinger, Daniel and Stringer, Carsen and Pachitariu, Marius},\n  journal = {Nature},\n  year    = {2025},\n  doi     = {10.1038/s41586-025-09180-y}\n}\n```\n\n## Source\n\n| | |\n|---|---|\n| Original dataset | Figshare (Janelia) article `28811129`, version 2 |\n| DOI | `10.25378/janelia.28811129.v2` |\n| Published | 2025-05-27 |\n| License | CC BY 4.0 |\n\nThis copy is redistributed under the same CC BY 4.0 licence as the original.\n\n## What was changed\n\n**Only the number of neurons per session.** The published release is 412 GB as\ndownloaded; this copy keeps a uniform random 9.8% of cells in each session,\nbringing it to 47 GB.\n\n| | full release | this copy |\n|---|---|---|\n| Mice | 19 | **19** (all) |\n| Sessions | 89 | **89** (all) |\n| Trials | 38,110 | **38,110** (all) |\n| Cells | 4,691,034 | **459,723** (9.8%) |\n\nNothing else is reduced. No mouse, session or trial is dropped, and no *type* of\ndata is removed — every behavioural variable, signal and auxiliary array in the\nfull release is still present. This matters for anyone comparing against\npublished statistics: per-session cell counts will be about a tenth of those\nreported, while animal, session and trial counts should match exactly.\n\nSubsampling is uniform random over the concatenated cell index, with a\nper-session seed derived from the session name, so the selection is deterministic\nand reproducible. `DATALIMIT_SUBSET.csv` records the exact counts:\n\n```\nsession,n_cells_total,n_cells_kept\nDR10_2022_07_12_1,58224,5706\n...\n```\n\nThe same cell index is applied to the corresponding `retinotopy/*_trans.npz`\narrays, so cell identity stays aligned across files.\n\n## Layout\n\n```\nbeh/          28 files, 6.6 GB    behavioural data, unmodified\nspk/          89 files, 40 GB     deconvolved activity, cells subsampled\nretinotopy/   89 files, 17 MB     retinotopic maps, same cell subset applied\nDATALIMIT_SUBSET.csv              per-session cell counts\n```\n\n`spk/<session>_neural_data.npy` is a 0-d pickled object array holding\n`{'spks': [plane0, plane1, plane2]}`; load with\n`np.load(path, allow_pickle=True).item()`.\n\n## Intended use\n\nBuilt for a data-reuse benchmark task, where an agent must reformat the published\ndata for a downstream decoder by working out the correct processing from the\npaper and reference code. The reduction keeps every source of variability that\ntask depends on — animals, sessions, trials, and all data types — while removing\nonly the redundancy of very large cell populations.\n\nFor any scientific analysis, use the full release from the DOI above rather than\nthis copy.\n","encoding":"utf8","key":"README.md","name":"README.md","size":4009,"truncated":false}],"sourceObjectKey":"DATALIMIT_SUBSET.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/kristinbranson/neurodata-reuse-zhong2025/DATALIMIT_SUBSET.csv\")\ndf.head()","starterSamplePath":"DATALIMIT_SUBSET.csv","totalObjects":209},"summary":"A size-reduced copy of the two-photon calcium imaging dataset from Zhong, Baptista, Gattoni, Arnold, Flickinger, Stringer and Pachitariu, Unsupervised pretraining in biological…","title":"Unsupervised Pretraining in Biological Neural Networks: Mouse Visual Cortex Two-Photon Calcium Imaging of 19 Mice, Neuron-Subsampled Copy (Zhong, Pachitariu, Janelia)","updatedAt":"2026-11-22T04:04:16.832Z","visibility":"public","websiteUrl":"https://doi.org/10.25378/janelia.28811129.v2"}}