---
pretty_name: BFD Source Archive Index
license: other
tags:
- biology
- protein
- sequence-database
- bfd
- fasta
- archive
- parquet
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*.parquet
  - split: test
    path: data/test-*.parquet
---

# BFD Source Archive Index

BFD is a very large clustered protein sequence database built from UniProt and metagenomic sequence resources, commonly used for homology search and multiple sequence alignment generation.

## Splits

The split is deterministic by chunk identifier: `sha256(index_id) % 10`. Bucket `0` is `test`; buckets `1` through `9` are `train`.

| Split | Rows |
|---|---:|
| train | 255 |
| test | 17 |
| total | 272 |

## Source File

| File | Size |
|---|---:|
| `bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt.tar.gz` | 291,649,557,441 bytes |

## Usage

```bash
pip install datasets
```

Load the archive index:

```python
from datasets import load_dataset

ds = load_dataset("LiteFold/BFD")
print(ds)
print(ds["train"][0])
```

Load one split:

```python
from datasets import load_dataset

train = load_dataset("LiteFold/BFD", split="train")
test = load_dataset("LiteFold/BFD", split="test")
```

Download the original archive:

```python
from huggingface_hub import hf_hub_download

archive_path = hf_hub_download(
    repo_id="LiteFold/BFD",
    repo_type="dataset",
    filename="bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt.tar.gz",
)
```

Load the source-file manifest:

```python
import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="LiteFold/BFD",
    repo_type="dataset",
    filename="metadata/source_files.parquet",
)
source_files = pd.read_parquet(path)
print(source_files)
```

## Columns

| Column | Description |
|---|---|
| `index_id` | Stable row ID for one compressed chunk. |
| `repo_id` | Hugging Face dataset repository. |
| `source_file` | Source archive filename. |
| `source_sha` | Source repository commit used to build the index. |
| `source_format` | Source archive format. |
| `chunk_index` | Zero-based compressed chunk index. |
| `byte_start` | Inclusive compressed byte offset. |
| `byte_end_exclusive` | Exclusive compressed byte offset. |
| `chunk_size_bytes` | Compressed chunk size in bytes. |
| `total_size_bytes` | Full compressed archive size. |
| `chunk_size_gib` | Target chunk size in GiB. |
| `is_first_chunk` | Whether the row indexes the first chunk. |
| `is_last_chunk` | Whether the row indexes the final chunk. |
| `access_note` | Note describing the index scope. |
| `split_bucket` | Deterministic split bucket from `sha256(index_id) % 10`. |

# Citation

```
@article{steinegger2019bfd,
  title   = {Protein-level assembly increases protein sequence recovery from metagenomic samples manyfold},
  author  = {Steinegger, Martin and Mirdita, Milot and S{\"o}ding, Johannes},
  journal = {Nature Methods},
  volume  = {16},
  pages   = {603--606},
  year    = {2019},
  doi     = {10.1038/s41592-019-0437-4}
}
```