IPknot++

IPknot++ is a benchmark dataset released with the IPknot++ RNA secondary structure prediction paper. It is intended as an external evaluation set rather than a training split. A common use is to train or tune RNA secondary structure prediction models on SPOT-RNA-style splits such as bpRNA-spot, then evaluate generalization on IPknot++.

The original release does not define train and validation partitions, so every variant is released as a single test split.

IPknot++ is released as three repositories that share the same RNA targets but differ in whether and how a multiple sequence alignment is provided:

• ipknot_plus_plus: single-sequence structures with no alignment, covering both the bpRNA-1m and Rfam 14.5 targets.
• ipknot_plus_plus-ref: Rfam 14.5 targets with the Rfam reference alignment, for common-structure prediction.
• ipknot_plus_plus-mafft: Rfam 14.5 targets with a MAFFT alignment, for common-structure prediction.

The -ref and -mafft variants cover the same set of Rfam 14.5 targets, with identical id, sequence, and secondary_structure; only aligned_sequences differs (the alignment members in aligned_ids are also identical). These targets are a subset of the single sequences in the no-alignment ipknot_plus_plus variant.

For the alignment variants, sequence and secondary_structure are taken from the corresponding single-sequence BPSEQ file, and the full alignment is stored in aligned_ids and aligned_sequences.

Schema

The ipknot_plus_plus variant contains:

Column	Description
id	Identifier of the benchmark entry.
sequence	Ungapped RNA sequence for the benchmark target.
secondary_structure	Target secondary structure in dot-bracket notation.

The ipknot_plus_plus-ref and ipknot_plus_plus-mafft variants add two alignment columns:

Column	Description
aligned_ids	Sequence IDs in the alignment. The first entry is the benchmark target itself.
aligned_sequences	Aligned sequences, preserving gap characters.

Disclaimer

This is an UNOFFICIAL release of the IPknot++ benchmark dataset by Kengo Sato and Yuki Kato.

The team releasing IPknot++ did not write this dataset card for this dataset so this dataset card has been written by the MultiMolecule team.

Dataset Description

• Homepage: https://www.sato-lab.org/en/publication/sato-2022-qk/
• datasets: https://huggingface.co/datasets/multimolecule/ipknot_plus_plus
• Original URL: https://zenodo.org/records/4923158

License

The original IPknot++ benchmark dataset is licensed under the Creative Commons Attribution 4.0 International license.

SPDX-License-Identifier: CC-BY-4.0

Citation

@article{sato2022prediction,
  author    = {Sato, Kengo and Kato, Yuki},
  journal   = {Briefings in Bioinformatics},
  month     = jan,
  number    = 1,
  title     = {Prediction of {RNA} secondary structure including pseudoknots for long sequences},
  volume    = 23,
  year      = 2022
}

@dataset{sato2021ipknot,
  author    = {Sato, Kengo and Kato, Yuki},
  title     = {{IPknot++} benchmark dataset},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.4923158},
  year      = 2021
}

Note

The artifacts distributed in this repository are part of the MultiMolecule project. If MultiMolecule supports your research, please cite the MultiMolecule project as follows:

@software{chen_2024_12638419,
  author    = {Chen, Zhiyuan and Zhu, Sophia Y.},
  title     = {MultiMolecule},
  doi       = {10.5281/zenodo.12638419},
  publisher = {Zenodo},
  url       = {https://doi.org/10.5281/zenodo.12638419},
  year      = 2024,
  month     = may,
  day       = 4
}