# Observed Antibody Space (OAS) unpaired heavy and light chains, Parquet

## Facts

- Provider: Oxford Protein Informatics Group (OAS); Parquet conversion by Converge Bio
- License: CC-BY-4.0 (https://opig.stats.ox.ac.uk/webapps/oas/)
- Size: 4,032 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/ConvergeBio/oas-unpaired
- Access: public, readable without an account
- Category: Biology
- Source: https://doi.org/10.1002/pro.4205
- Page: https://scigantic.com/archive/73b16e27-5510-5673-912b-fb2c44aea3ef
- Keywords: Biology, Parquet, Hugging Face

## About

The unpaired portion of the Observed Antibody Space (OAS), the Oxford Protein Informatics Group's database of cleaned, annotated antibody repertoire sequencing data, converted to Parquet by Converge Bio. It holds about 2.07 billion heavy-chain and 357 million light-chain sequences, about 2.7 TB in total. Each row carries the AIRR-style annotation OAS provides, including the aligned amino-acid sequence, CDR3, V and J gene calls, and study and subject metadata. The repo exposes configs for all heavy chains, all light chains, and one config per source study (for example Briney et al., 2019), so a single study or a column subset can be read without touching the rest. OAS is the standard pretraining corpus for antibody language models. OAS data is CC BY 4.0; cite Olsen, Boyles and Deane, Protein Science 2022.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("ConvergeBio/oas-unpaired")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/unpaired_heavy/Banerjee et al., 2017/train-00000.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/73b16e27-5510-5673-912b-fb2c44aea3ef
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/73b16e27-5510-5673-912b-fb2c44aea3ef
- Over MCP: `get_schema_card` with id `73b16e27-5510-5673-912b-fb2c44aea3ef` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Biology datasets](https://scigantic.com/topics/biology)
