DeepFoldProtein/malidup-dataset

MALIDUP (Hugging Face Port)

Benchmark of duplicated domain pairs from the same protein chain, curated by structure-based alignment.

Overview

• 241 duplicated domain pairs from the same polypeptide chains
• Expert-curated alignments based on structural similarity
• Useful for evaluating alignment methods under remote but homologous scenarios

Features

• pair_id, group_id, set_name
• seq1_id, seq2_id, seq1, seq2
• ref_alignment: list of aligned residue index pairs (0-based)
• percent_identity: sequence identity across aligned positions
• scop_labels: SCOP class, fold, superfamily, family, etc.
• meta: additional metadata (JSON string)

Usage

from datasets import load_dataset
ds = load_dataset("DeepFoldProtein/MALIDUP", name="all", split="test")
ex = ds[0]
print(ex["seq1_id"], ex["seq2_id"], ex["ref_alignment"][:5])

Citation

Please cite:

@article{Cheng2007,
  title = {MALIDUP: A database of manually constructed structure alignments for duplicated domain pairs},
  volume = {70},
  ISSN = {1097-0134},
  url = {http://dx.doi.org/10.1002/prot.21783},
  DOI = {10.1002/prot.21783},
  number = {4},
  journal = {Proteins: Structure,  Function,  and Bioinformatics},
  publisher = {Wiley},
  author = {Cheng, Hua and Kim, Bong‐Hyun and Grishin, Nick V.},
  year = {2007},
  month = oct,
  pages = {1162--1166}
}