---
license: apache-2.0
language:
- en
configs:
- config_name: alcohol_100_bacteria_0
  data_files:
  - split: train
    path: alcohol_100_bacteria_0/train.csv
  - split: validation
    path: alcohol_100_bacteria_0/valid.csv
  - split: id_test
    path: alcohol_100_bacteria_0/id_test.csv
  - split: ood_test
    path: alcohol_100_bacteria_0/ood_test.csv
- config_name: alcohol_75_bacteria_25
  data_files:
  - split: train
    path: alcohol_75_bacteria_25/train.csv
  - split: validation
    path: alcohol_75_bacteria_25/valid.csv
  - split: id_test
    path: alcohol_75_bacteria_25/id_test.csv
  - split: ood_test
    path: alcohol_75_bacteria_25/ood_test.csv
- config_name: alcohol_50_bacteria_50
  data_files:
  - split: train
    path: alcohol_50_bacteria_50/train.csv
  - split: validation
    path: alcohol_50_bacteria_50/valid.csv
  - split: id_test
    path: alcohol_50_bacteria_50/id_test.csv
  - split: ood_test
    path: alcohol_50_bacteria_50/ood_test.csv
- config_name: alcohol_0_bacteria_100
  data_files:
  - split: train
    path: alcohol_0_bacteria_100/train.csv
  - split: validation
    path: alcohol_0_bacteria_100/valid.csv
  - split: id_test
    path: alcohol_0_bacteria_100/id_test.csv
  - split: ood_test
    path: alcohol_0_bacteria_100/ood_test.csv
task_categories:
- text2text-generation
- text-generation
pretty_name: Alcohol and Bacteria Metadata Harmonization Dataset
---

# Alcohol and Bacteria Metadata Harmonization Dataset

## Summary

This dataset contains domain-specific term mixtures for training and evaluating metadata harmonization systems under domain shift. Each configuration includes a defined ratio of alcohol-related and bacteria-related terms to support experiments on generalization and domain adaptation. Each entry includes a term representation, its corresponding harmonized standard, and metadata such as variation type and source terminology. Variation types range from standard forms to lexical (e.g., synonyms, abbreviations) and structural modifications (e.g., removing punctuation or grouping symbols, omitting stopwords, or rearranging word order). Term harmonization targets are drawn from PhenX Toolkit, NCBI Taxonomy, and NIST.

This dataset was used in the study:
> **Metadata Harmonization from Biological Datasets with Language Models**  
> Alexander Verbitsky, Patrick Boutet, Mohammed Eslami  
> Netrias, LLC

## Configurations

Each configuration provides a different mixture of alcohol and bacteria terms for harmonization:

- `alcohol_100_bacteria_0`: 100% alcohol-related terms
- `alcohol_75_bacteria_25`: 75% alcohol, 25% bacteria
- `alcohol_50_bacteria_50`: 50% alcohol, 50% bacteria
- `alcohol_25_bacteria_75`: 25% alcohol, 75% bacteria
- `alcohol_0_bacteria_100`: 100% bacteria-related terms

## Supported Splits

Each configuration includes the following splits:

- `train`: Training data with aligned representation-standard pairs.
- `validation`: Used for model selection and early stopping.
- `id_test`: In-dictionary test set; contains new representations of standard terms present in the training set.
- `ood_test`: Out-of-dictionary test set; contains representations of standard terms not present in the training set.

## Features

Each example includes the following fields:

- `representation` (*string*): The original term variant being harmonized. This may be a synonym, abbreviation, transformation, or the standard term itself.
- `harmonization_standard` (*string*): The correct standardized form of the term, drawn from a controlled terminology.
- `variation_type` (*string*): A label describing how the representation differs from the standard (e.g., “abbreviation”, “synonym”, or other transformation types).
- `source_terminology` (*string*): Comma-separated ontologies or terminologies where the standard term appears. Multiple sources are listed if the term is shared across different standards.

## Usage

```python
from datasets import load_dataset

# Load the 50% alcohol, 50% bacteria configuration
dataset = load_dataset("netrias/alcohol_bacteria_metadata_harmonization", "alcohol_50_bacteria_50")

train = dataset["train"]
validation = dataset["validation"]
id_test = dataset["id_test"]
ood_test = dataset["ood_test"]

print(f"Train Example:\n{train[0]}")
print(f"\nValidation Example:\n{validation[0]}")
print(f"\nIn-Dictionary Test Example:\n{id_test[0]}")
print(f"\nOut-of-Dictionary Test Example:\n{ood_test[0]}")
```

## Citation

If you use this dataset, please cite:

> Verbitsky A., Boutet P., Eslami M. (2025). *Metadata Harmonization from Biological Datasets with Language Models*. bioRxiv. https://doi.org/10.1101/2025.01.15.633281

## License

This dataset is released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0).