DeepFoldProtein/malisam-dataset

MALISAM (Hugging Face Port)

Benchmark for structural analogs—non-homologous protein motifs with similar 3D structures.

Overview

• 130 structurally aligned domain pairs with no evolutionary relationship
• First benchmark to explicitly evaluate analog discrimination
• Alignments are manually curated based on structural similarity alone

Features

• pair_id, group_id, set_name
• seq1_id, seq2_id, seq1, seq2
• ref_alignment: list of residue index pairs (0-based)
• percent_identity: sequence identity across aligned residues
• scop_labels: optional (typically empty for MALISAM)
• meta: optional notes extracted from original text files

Usage

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

Citation

@article{Cheng2007MALISAM,
  title = {MALISAM: a database of structurally analogous motifs in proteins},
  volume = {36},
  ISSN = {1362-4962},
  url = {http://dx.doi.org/10.1093/nar/gkm698},
  DOI = {10.1093/nar/gkm698},
  number = {Database},
  journal = {Nucleic Acids Research},
  publisher = {Oxford University Press (OUP)},
  author = {Cheng, H. and Kim, B.-H. and Grishin, N. V.},
  year = {2007},
  month = dec,
  pages = {D211--D217}
}