{
  "task_id": "human-mouse-clearance-expansionrx",
  "public_retrospective": true,
  "source": {
    "repository": "openadmet/openadmet-expansionrx-challenge-data",
    "revision": "6b898ccc43d10d25b230fb09e22a6e30c30022b5",
    "retrieved_date": "2026-10-14",
    "files": [
      {
        "filename": "README.md",
        "url": "https://huggingface.co/datasets/openadmet/openadmet-expansionrx-challenge-data/resolve/6b898ccc43d10d25b230fb09e22a6e30c30022b5/README.md",
        "sha256": "cb8a82b1414ab82bce1a50822d916327a414ea51e5cb129a01c855621e3d33d8",
        "size_bytes": 4388
      },
      {
        "filename": "expansion_data_train.csv",
        "url": "https://huggingface.co/datasets/openadmet/openadmet-expansionrx-challenge-data/resolve/6b898ccc43d10d25b230fb09e22a6e30c30022b5/expansion_data_train.csv",
        "sha256": "c5214ad8c8a4d7d4d09082fcce24fe16c97a980a1e0873b91b8c8e474b79f6e4",
        "size_bytes": 464076
      },
      {
        "filename": "expansion_data_test.csv",
        "url": "https://huggingface.co/datasets/openadmet/openadmet-expansionrx-challenge-data/resolve/6b898ccc43d10d25b230fb09e22a6e30c30022b5/expansion_data_test.csv",
        "sha256": "e21c0ef19795317216287e924e00ca3ef9388b7032a7fa4c7f27e2580372c85e",
        "size_bytes": 221291
      },
      {
        "filename": "expansion_data_raw.csv",
        "url": "https://huggingface.co/datasets/openadmet/openadmet-expansionrx-challenge-data/resolve/6b898ccc43d10d25b230fb09e22a6e30c30022b5/expansion_data_raw.csv",
        "sha256": "f674ec74cca1146bc386f832a32d4b8d921d3c312f92cb436cc005901c724a3c",
        "size_bytes": 686791
      }
    ]
  },
  "license": "CC-BY-4.0",
  "attribution": "OpenADMET and Expansion Therapeutics",
  "license_url": "https://creativecommons.org/licenses/by/4.0/",
  "transformation": {
    "train_file": "expansion_data_train.csv",
    "evaluation_files": [
      "expansion_data_test.csv"
    ],
    "column_mapping": {
      "HLM CLint": "hlm_clint",
      "MLM CLint": "mlm_clint"
    },
    "identifiers": "Molecule Name copied to molecule_id; SMILES copied byte-for-byte as field strings; source row order retained.",
    "labels": "All source numeric cell strings retained. Empty fields remain empty; no imputation or censor-bound substitution.",
    "evaluation": "Original released temporal test split retained; no raw censored labels used."
  },
  "train_rows": 5326,
  "evaluation_rows": 2282,
  "statistics": {
    "hlm_clint": {
      "train_observed": 3759,
      "train_missing": 1567,
      "train_reported_zero": 164,
      "evaluation_observed": 782,
      "evaluation_missing": 1500,
      "evaluation_reported_zero": 15,
      "training_constant": 16.4,
      "constant_baseline_mae": 0.3817467422385945
    },
    "mlm_clint": {
      "train_observed": 4522,
      "train_missing": 804,
      "train_reported_zero": 147,
      "evaluation_observed": 1170,
      "evaluation_missing": 1112,
      "evaluation_reported_zero": 9,
      "training_constant": 219.64999433491963,
      "constant_baseline_mae": 0.44403656223554705
    }
  },
  "baseline": {
    "method": "Per-endpoint training median in the scoring space",
    "score": 0.4128916522370708,
    "novel_model": false
  },
  "exact_id_overlap": 0,
  "exact_smiles_string_overlap": 0,
  "overlap_limitation": "No chemical canonicalization was used. Exact-string checks do not rule out salt, tautomer, stereo or chemical-series overlap.",
  "reference_hashes": {
    "labels.csv": "d89af23b9041387a836ef0087fe1e8b4c7c8e3226bd3428992932e334283e354",
    "test.csv": "1d7d86b3de02b26210d450520e6f8eefa6282f747e94c6ac67f8f6cb1bbe8652",
    "train.csv": "5bc5a5619f7fcb349e43199364f0e2abec17e93129086d1fc37c9c610872052e"
  },
  "raw_audit": "Every nonmissing HLM/MLM value in these train/test splits numerically equals its raw-file counterpart. Reported zeros are literal 0 in raw. Censor modifiers are not scored or substituted.",
  "source_caveat": "The upstream card names LICENSE.md, but the pinned repository tree does not contain that file; its card explicitly declares CC-BY-4.0. Attribution and license URL are retained here."
}
