pankajrajdeo/bond-czi-benchmark

BOND CZI Benchmark Dataset

A comprehensive benchmark dataset for ontology mapping in biomedical single-cell data, derived from CELLxGENE metadata.

Dataset Description

This dataset contains 192,916 training examples of author terms mapped to standardized ontology identifiers across 7 field types: assay, cell_type, development_stage, disease, self_reported_ethnicity, sex, and tissue.

Data Structure

Each example contains:

• Dataset context: dataset_id, dataset_title, collection_name
• Anchor context: field_type, organism, tissue, author_term
• Original ontology mapping: ontology_id, label, definition, synonyms, obsolete status
• Resolved ontology mapping: resolved obsolete terms with full metadata
• Evidence metrics: support counts and confidence scores

Splits

• Train: 173,632 examples (90%)
• Validation: 9,639 examples (5%)
• Test: 9,645 examples (5%)

Field Types

1. assay: Experimental methods and technologies
2. cell_type: Cell classifications and types
3. development_stage: Developmental stages and ages
4. disease: Disease conditions and phenotypes
5. self_reported_ethnicity: Ethnicity and ancestry information
6. sex: Biological sex classifications
7. tissue: Tissue and organ types

Data Format

The dataset is provided in JSONL format with the following structure:

{
  "dataset_id": "00ff600e-6e2e-4d76-846f-0eec4f0ae417",
  "dataset_title": "Human tonsil nonlymphoid cells scRNA",
  "collection_name": "Single-cell analysis of human B cell maturation...",
  "field_type": "cell_type",
  "organism": "Homo sapiens",
  "tissue": "tonsil",
  "author_term": "cDC1",
  "original_ontology_id": "CL:0000990",
  "original_is_obsolete": 0,
  "original_label": "conventional dendritic cell",
  "original_definition": "Conventional dendritic cell is a dendritic cell...",
  "original_synonyms_exact": "DC1|cDC|dendritic reticular cell|type 1 DC",
  "original_synonyms_narrow": "",
  "original_synonyms_broad": "interdigitating cell|veiled cell",
  "original_synonyms_related": "interdigitating cell|veiled cell|DC1|cDC|dendritic reticular cell|type 1 DC",
  "original_replaced_by": "",
  "original_consider": "",
  "resolved_ontology_id": "CL:0000990",
  "resolved_label": "conventional dendritic cell",
  "resolved_definition": "Conventional dendritic cell is a dendritic cell...",
  "resolved_synonyms_exact": "DC1|cDC|dendritic reticular cell|type 1 DC",
  "resolved_synonyms_narrow": "",
  "resolved_synonyms_broad": "interdigitating cell|veiled cell",
  "resolved_synonyms_related": "interdigitating cell|veiled cell|DC1|cDC|dendritic reticular cell|type 1 DC",
  "resolution_path": "CL:0000990",
  "match_status": "unchanged",
  "support_dataset_count": 2,
  "support_row_count": 56,
  "llm_predicted_author_column": "author_cell_type",
  "author_confidence": 0.98,
  "ontology_confidence": 1.0,
  "split": "train"
}

Note: Array fields (synonyms, replaced_by, consider, resolution_path) are stored as pipe-separated strings for better Hugging Face compatibility. Empty arrays are represented as empty strings.

Usage

This dataset is designed for training models to map free-text author terms to standardized ontology identifiers in biomedical contexts. It can be used for:

• Ontology mapping and entity linking
• Biomedical named entity recognition
• Text classification in biomedical domains
• Training biomedical language models

Files

• bond_czi_benchmark_data_hf_train.jsonl: Training split (173,632 examples)
• bond_czi_benchmark_data_hf_dev.jsonl: Validation split (9,639 examples)
• bond_czi_benchmark_data_hf_test.jsonl: Test split (9,645 examples)

Citation

If you use this dataset, please cite the original CELLxGENE publications and the BOND project.