ConvergeBio/uniclust30

---
language: en
license: cc-by-sa-4.0
tags:
- biology
- protein
- protein-sequences
- uniclust
- uniclust30
- uniref30
- msa
- multiple-sequence-alignment
- proteomics
- bioinformatics
pretty_name: UniClust30 (UniRef30)
size_categories:
- 10M<n<100M
task_categories:
- feature-extraction
configs:
- config_name: default
data_files:
- split: train
path: "train-*.parquet"
dataset_info:
features:
- name: cluster_id
dtype: string
- name: representative_id
dtype: string
- name: sequence
dtype: large_string
- name: sequence_length
dtype: int32
- name: sequence_xxh128
dtype: string
- name: num_aligned
dtype: int32
- name: a3m
dtype: large_string
- name: member_count
dtype: int32
- name: member_ids
sequence: string
splits:
- name: train
num_examples: 36293491
---

# UniClust30 (UniRef30)

Complete [UniClust30 / UniRef30](https://uniclust.mmseqs.com/) dataset (release 2023_02) from the S&ouml;ding Lab, converted from HH-suite A3M format to sharded Parquet. UniClust30 clusters UniProt sequences at 30% identity and includes precomputed multiple sequence alignments (MSAs) &mdash; widely used as input for protein structure prediction (AlphaFold, ColabFold) and protein language model pretraining.

**Part of the [ConvergeBio Protein Database Collection](https://huggingface.co/collections/ConvergeBio/protein-database)** &mdash; see also [UniRef50](https://huggingface.co/datasets/ConvergeBio/uniref50), [UniRef90](https://huggingface.co/datasets/ConvergeBio/uniref90), and [UniRef100](https://huggingface.co/datasets/ConvergeBio/uniref100).

## Dataset Summary

| | |
|---|---|
| **Clusters** | 36,293,491 |
| **Shards** | 629 |
| **Release** | 2023_02 |
| **Includes** | Precomputed A3M multiple sequence alignments per cluster |

## Schema

Each row represents one UniClust30 cluster with its representative sequence, MSA, and membership information.

| Column | Type | Description |
|--------|------|-------------|
| `cluster_id` | `string` | Cluster identifier (UniRef30 accession) |
| `representative_id` | `string` | UniProt accession of the representative sequence |
| `sequence` | `large_string` | Representative protein sequence (uppercase amino acid alphabet) |
| `sequence_length` | `int32` | Length of the representative sequence in residues |
| `sequence_xxh128` | `string` | xxHash-128 of the sequence (hex, computed at build time) |
| `num_aligned` | `int32` | Number of sequences in the A3M multiple sequence alignment |
| `a3m` | `large_string` | Full A3M-formatted MSA for the cluster |
| `member_count` | `int32` | Number of cluster members (from mapping file) |
| `member_ids` | `list<string>` | All member UniProt accessions |

## Usage

```python
from datasets import load_dataset

# Stream without downloading everything
ds = load_dataset("ConvergeBio/uniclust30", streaming=True)
for row in ds["train"]:
print(row["cluster_id"], row["sequence_length"], row["num_aligned"])
break

# Or load fully
ds = load_dataset("ConvergeBio/uniclust30")
```

## Data Processing

- **Source:** HH-suite ffindex/ffdata A3M database and `uniref_mapping.tsv.gz` from [uniclust.mmseqs.com](https://uniclust.mmseqs.com/)
- **Parsing:** Direct ffindex/ffdata binary reads; membership from mapping TSV
- **Integrity:** xxHash-128 computed per sequence; A3M representative sequence verified against extracted sequence
- **Validation:** Passed all tiers &mdash; schema conformance, zero null/empty sequences, xxHash roundtrip, A3M format checks, A3M&ndash;sequence consistency, member ID mapping verification, and field-by-field comparison against source ffindex/ffdata
- **Format:** Sharded Parquet with zstd compression

## Source & Citation

UniClust30 is produced by the [S&ouml;ding Lab](https://www.mpinat.mpg.de/soeding):

> Mirdita M, von den Driesch L, Galiez C, Martin MJ, S&ouml;ding J, Steinegger M.
> "Uniclust databases of clustered and deeply annotated protein sequences and
> alignments." *Nucleic Acids Res.* 45(D1):D170&ndash;D176 (2017).
> [doi:10.1093/nar/gkw1081](https://doi.org/10.1093/nar/gkw1081)

## About

Built by [Converge Bio](https://converge-bio.com) &mdash; accelerating drug discovery with generative AI. Converge Bio develops foundation models for protein engineering, antibody design, and gene expression optimization, powering its computational lab products ConvergeAB, ConvergeGEO, and ConvergeCELL.

## License

UniClust30 data is available under
[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).