Protein Sequences · LiteFold
BFD
An indexed distribution of the BFD clustered protein sequence database for homology search and alignment.
What BFD is
BFD is a large clustered protein sequence database compiled from UniProt and metagenomic sequence resources. It serves as a standard reference sequence collection for homology searching and for assembling multiple sequence alignments (MSAs), notably within AlphaFold-style structural biology workflows.
The LiteFold repository provides an index over the core compressed archive, bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt.tar.gz.
How it was made
The repository indexes a single primary archive file totalling 291,649,557,441 bytes into 272 structured chunks. Each entry in the Parquet index specifies positional markers including byte_start, byte_end_exclusive, chunk_size_bytes, and chunk_size_gib, alongside flags denoting whether an entry is the first or last chunk of the archive.
Dataset splits are generated deterministically using the formula sha256(index_id) % 10. Chunks hashing to bucket 0 are assigned to the test split (17 rows), while buckets 1 through 9 form the train split (255 rows). File tracking metadata is additionally provided in a dedicated Parquet manifest.
Key facts
- Publisher
- LiteFold
- Category
- Protein Sequences
- Primary source file
- bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt.tar.gz
- Archive size
- 291,649,557,441 bytes
- Total index rows
- 272
- Train split rows
- 255
- Test split rows
- 17
- Index format
- Parquet
What you can use it for
- Homology search
- Comparing target protein sequences against clustered metagenomic and UniProt sequences to detect distant evolutionary relatives.
- Multiple sequence alignment generation
- Constructing alignments across sequence families to serve as input features for protein structure prediction models such as AlphaFold.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from LiteFold/BFD at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 255 rows of default/train
access_note
string_label- Compressed byte-range index for the BFD source archive; download or stream the original tar.gz for sequence records.100%
byte_end_exclusive
int- min
- 1.1B
- median
- 142.8B
- mean
- 143.9B
- max
- 289.9B
byte_start
int- min
- 0
- median
- 141.7B
- mean
- 142.8B
- max
- 288.8B
chunk_index
int- min
- 0
- median
- 132
- mean
- 132.99
- max
- 269
chunk_size_bytes
int- min
- 1.1B
- median
- 1.1B
- mean
- 1.1B
- max
- 1.1B
chunk_size_gib
int- min
- 1
- median
- 1
- mean
- 1
- max
- 1
index_id
string_text- min
- 76
- median
- 76
- mean
- 76
- max
- 76
is_first_chunk
bool- False100%
- True0%
is_last_chunk
bool- False100%
repo_id
string_label- LiteFold/BFD100%
source_file
string_label- bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt.tar.gz100%
source_format
string_label- tar.gz100%
source_sha
string_label- 61cd373acbdc6265b698ea70bd88c8dd8f0b4d27100%
split_bucket
int- min
- 1
- median
- 5
- mean
- 4.87
- max
- 9
total_size_bytes
int- min
- 291.6B
- median
- 291.6B
- mean
- 291.6B
- max
- 291.6B
Strengths and limits
Good at
- Combines both UniProt and diverse metagenomic data into a clustered sequence reference.
- Splits are generated deterministically via SHA-256 hashing on chunk identifiers.
Watch out for
- Storing the full unpartitioned sequence file requires 291,649,557,441 bytes of storage.
- The dataset consists of archive indexing metadata rather than pre-parsed individual protein records.
Questions
Can I look inside BFD without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is BFD under?
Its authors declare other. Read the terms on the original page before reusing the data.
What sequence sources make up BFD?
BFD is composed of clustered protein sequences drawn from UniProt and metagenomic sequence resources.
What is the size of the underlying sequence archive?
The underlying archive, bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt.tar.gz, is 291,649,557,441 bytes in size.
What information does each chunk index record contain?
Each row provides the chunk index, inclusive start byte, exclusive end byte, chunk size in bytes and gibibytes, total archive size, boolean chunk position flags, and repository source identifiers.
How are the train and test splits partitioned?
Rows are split based on sha256(index_id) % 10, placing bucket 0 into the test split (17 chunks) and buckets 1 to 9 into the train split (255 chunks).
How to cite it
Steinegger, Martin and Mirdita, Milot and S{"o}ding, Johannes. Protein-level assembly increases protein sequence recovery from metagenomic samples manyfold. Nature Methods 16, 603--606 (2019). doi: 10.1038/s41592-019-0437-4
Written from BFD's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.