Protein Families · LiteFold
Pfam
A JSONL mirror of Pfam-A protein domain annotations and clan mappings with row-level provenance.
What Pfam is
Pfam is a reference biological resource providing protein family and domain classifications derived from curated multiple sequence alignments and profile hidden Markov models (profile HMMs). This dataset provides a normalised mirror of tabular exports from the Pfam current release, packaged in JSON Lines (JSONL) format. It aims to support large-scale domain annotation queries, protein family investigations, model training over domain coordinates, and linking with protein sequence or structure datasets.
The mirror contains two principal Pfam-A tables: a clan table and a sequence regions table. The clan table associates individual Pfam-A families with wider evolutionary groupings known as clans and includes family identifiers and names. The regions table records specific residue coordinates where Pfam-A domains occur across known protein sequences. Each record embeds the upstream tabular data alongside explicit provenance fields to trace the source file and row index.
How it was made
The dataset structures upstream Pfam release tables into JSONL records containing four outer keys: the dataset identifier, the original source file path, the zero-based row index in that source file, and a nested row object containing the original data fields. The repository is partitioned into two distinct configurations: the clans configuration containing 27,480 records, and the regions configuration containing 128,192,943 records split across two sequential JSONL parts.
In the regions table, each record details a sequence-domain assignment, providing the protein accession, sequence version number, CRC64 and MD5 sequence checksums, Pfam-A accession, sequence boundaries (seq_start and seq_end), and alignment match boundaries (ali_start and ali_end). In the clans table, records contain the Pfam accession, family identifier, family name, clan accession, and clan name.
Key facts
- Total rows
- 128,220,423
- Clans split rows
- 27,480
- Regions split rows
- 128,192,943
- Format
- JSONL
- Upstream tables included
- Pfam-A clans, Pfam-A regions
What you can use it for
- Protein Family Representation Learning
- Train machine learning models on Pfam-A region assignments and sequence-domain boundary pairs.
- Protein Sequence Integration
- Cross-reference external sequence databases with Pfam domains using sequence accessions, sequence versions, CRC64 checksums, or MD5 checksums.
- Clan Hierarchy Mapping
- Map individual Pfam-A families to their broader clan groupings and family naming metadata.
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/Pfam at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 27,480 rows of clans/train
dataset_id
string_label- pfam100%
row_index
int- min
- 0
- median
- 13739.50
- mean
- 13739.50
- max
- 27,479
source_file
string_label- annotation/pfam/current_release/Pfam-A.clans.tsv.gz100%
Strengths and limits
Good at
- Provides row-level provenance for every record using original source file paths and line indices.
- Includes both sequence boundary coordinates and alignment coordinates alongside CRC64 and MD5 sequence checksums.
Watch out for
- Does not include complete Pfam distribution assets such as seed alignments, full multiple sequence alignments, or profile hidden Markov model files.
- The clans and regions tables have different nested row schemas and cannot be parsed together as a single unified table without schema errors.
Questions
Can I look inside Pfam 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 Pfam under?
Its authors declare cc0-1.0. Read the terms on the original page before reusing the data.
Does this dataset include Pfam profile HMMs or alignment files?
No. This repository is a normalised mirror of Pfam tabular exports and does not contain the profile hidden Markov models, seed alignments, or full alignments.
How is provenance tracked for each record?
Every JSON record contains the upstream Pfam file name in source_file and the zero-based index in row_index, alongside the nested row containing the original table fields.
Can the entire dataset be loaded at once?
Loading all JSONL files into a single unified dataset table can cause schema casting errors because the clans and regions tables have different nested row fields. They should be accessed via their respective configurations.
What checksums are available for sequence matching?
The regions table provides both CRC64 and MD5 sequence checksums for verifying protein sequence identity.
How to cite it
@article{mistry2021pfam, title = {{Pfam}: The protein families database in 2021}, author = {Mistry, Jaina and Chuguransky, Sara and Williams, Lowri and Qureshi, Matloob and Salazar, Gustavo A. and Sonnhammer, Erik L. L. and Tosatto, Silvio C. E. and Paladin, Lisanna and Raj, Shriya and Richardson, Lorna J. and Finn, Robert D. and Bateman, Alex}, journal = {Nucleic Acids Research}, volume = {49}, number = {D1}, pages = {D412--D419}, year = {2021}, doi = {10.1093/nar/gkaa913}, url = {https://doi.org/10.1093/nar/gkaa913}}
Written from Pfam'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.