Datasets:
The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 80, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
CATH 4.3 Chain-Level Dataset
Chain-level protein structures derived from the CATH 4.3 S40 non-redundant
dataset. CATH distributes structures as seven-character domain entries
(for example 2efiA00); this dataset provides five-character chain-level
entries (2efiA), together with the corresponding single-chain PDB files.
The dataset was assembled for ProteinZero but is released independently of it.
Contents
| Split | Structures |
|---|---|
| train | 22,761 |
| validation | 1,040 |
| test | 985 |
| total | 24,786 |
Structure identifiers do not overlap across the three splits. Sequence lengths range from 40 to 3,042 residues.
The archive contains:
metadata.csv— one row per structure- 15 reference partition files
pdbs/{train,val,test}/— one single-chain PDB file per structureDATASET.md— full documentation
metadata.csv
| Field | Definition |
|---|---|
structure_id |
Five-character chain identifier: four-character PDB accession followed by the chain identifier (case-sensitive) |
seq |
Amino-acid sequence |
length |
Number of residues in seq |
split |
train, val, or test |
Reference partitions
Each split provides one complete file and four disjoint length subsets:
{train,val,test}_data.csv all structures in the split
{train,val,test}_data_le100.csv at most 100 residues
{train,val,test}_data_101-300.csv 101-300 residues
{train,val,test}_data_301-500.csv 301-500 residues
{train,val,test}_data_ge501.csv at least 501 residues
The four subsets are mutually exclusive and their union equals the
corresponding complete file. Alternative partitions can be constructed
directly from the length field in metadata.csv.
Download
hf download ZiwenWang28/cath43-chain-level \
cath43_chain_dataset.tar.gz \
--repo-type=dataset --local-dir .
tar -xzf cath43_chain_dataset.tar.gz
Provenance
Derived from the CATH 4.3 S40 non-redundant set: https://download.cathdb.info/cath/releases/all-releases/v4_3_0/non-redundant-data-sets/
CATH domain records were aggregated to chain level, with full chains retrieved from RCSB where needed. Each PDB file contains only the chain identified by the fifth character of its filename. Chain identifiers are case-sensitive.
Notes on downstream use
Sequence-length constraints apply when this dataset is used with the InstructPLM pretrained model, as in ProteinZero. These are properties of that model, not of the dataset:
- above 512 residues, the structure encoder silently truncates the structure representation to its first 512 positions;
- above 766 residues, the model input (256 structure tokens, two sentinel tokens, and L residue tokens) exceeds the 1,024-token attention window;
- above 1,000 residues, the ESMFold-based reward path is skipped.
The dataset is released across its full observed length range to support use
beyond that codebase. See DATASET.md in the archive for details.
License
This dataset is released under CC-BY-4.0.
It is derived from CATH, which is licensed under CC-BY-4.0, and from the Protein Data Bank, whose data are in the public domain. Please cite all three references below.
Citation
If you use this dataset, please cite ProteinZero along with the original data sources.
@article{wang2025proteinzero,
title={Proteinzero: Self-improving protein generation via online reinforcement learning},
author={Wang, Ziwen and Fan, Jiajun and Guo, Ruihan and Nguyen, Thao and Ji, Heng and Liu, Ge},
journal={arXiv preprint arXiv:2506.07459},
year={2025}
}
@article{sillitoe2021cath,
title={CATH: increased structural coverage of functional space},
author={Sillitoe, Ian and Bordin, Nicola and Dawson, Natalie and Waman, Vaishali P and Ashford, Paul and Scholes, Harry M and Pang, Camilla SM and Woodridge, Laurel and Rauer, Clemens and Sen, Neeladri and others},
journal={Nucleic acids research},
volume={49},
number={D1},
pages={D266--D273},
year={2021},
publisher={Oxford University Press}
}
@article{berman2000protein,
title={The protein data bank},
author={Berman, Helen M and Westbrook, John and Feng, Zukang and Gilliland, Gary and Bhat, Talapady N and Weissig, Helge and Shindyalov, Ilya N and Bourne, Philip E},
journal={Nucleic acids research},
volume={28},
number={1},
pages={235--242},
year={2000},
publisher={Oxford University Press}
}
- Downloads last month
- -