{
  "READY": true,
  "name": "hnoca_cells30",
  "task": "classification",
  "num_classes": 30,
  "cat_idx": [],
  "num_samples": 100000,
  "num_features": 22000,
  "imbalance_ratio": 69.2599,
  "description": "Single-cell RNA-seq of human neural organoids from the HNOCA atlas (27 publications). Task: predict the HNOCA harmonized level-4 cell-type annotation (annot_level_4_rev2) among its 30 most frequent classes. The split is donor- and publication-disjoint: groups are connected components of the publication-cell line graph and are assigned to train/val/test (train: Bhaduri, 2020; Fiorenzano, 2021; Fleck, 2022; Huang, 2021; Kanton, 2019; Kelava, 2022; Paulsen, 2022; Pellegrini, 2020; Sawada, 2020; Uzquiano, 2022; Velasco, 2019; Xiang, 2019; val: Bowles, 2021; He, 2022; test: Andersen, 2020; Atamian, 2024; Birey, 2017; Khan, 2020; Marton, 2019; Miura, 2020; Qian, 2020; Samarasinghe, 2021; Sloan, 2017; Treutlein, 2023; Trujillo, 2019; Vertesy, 2022; Yoon, 2019); then 70000/15000/15000 cells are sampled within the splits, stratified by class. Features: log1p(CPM) expression of 22000 genes selected with analytic Pearson residuals (scanpy, flavor='pearson_residuals') computed on the raw counts of the training cells only.",
  "source": "He et al. (2024), Human Neural Organoid Cell Atlas (HNOCA), hnoca_cleaned.h5ad downloaded from https://zenodo.org/records/14161275.",
  "label": "Single-Cell",
  "sub_labels": [
    "RNA-seq"
  ],
  "license": "CC-BY-4.0",
  "license_url": "https://creativecommons.org/licenses/by/4.0/",
  "source_licenses": [
    {
      "id": "hnoca",
      "title": "HNOCA cleaned datasets, Zenodo version 2",
      "url": "https://zenodo.org/records/14161275",
      "license": "CC-BY-4.0",
      "license_url": "https://creativecommons.org/licenses/by/4.0/",
      "terms_url": null
    }
  ]
}