# Precomputed RNA MSA resource (RNAcentral × Riboseek)

A precomputed multiple-sequence-alignment (MSA) resource covering the
structured, non-redundant subset of RNAcentral, with one MSA per query
sequence generated by Riboseek.

  Riboseek:  https://github.com/steineggerlab/riboseek
  Preprint:  https://doi.org/10.64898/2026.07.31.741718

Accompanies Park et al., "Fast remote nucleotide sequence alignment with
Riboseek", bioRxiv (2026).


## Dataset construction

All sequences were downloaded from RNAcentral Release 25. Sequences containing
gap characters or nonstandard nucleotides were removed, followed by
exact-sequence deduplication.

Secondary structures were predicted with EternaFold (Wayment-Steele et al.,
2022) using default settings. Sequences were retained when at least 60% of
nucleotides were predicted to participate in base pairing; sequences of 600
nucleotides or longer were excluded. This yielded 1,731,677 query sequences.

Riboseek was then used to generate one MSA per retained sequence, following the
same two-stage procedure as in the secondary-structure benchmark (see
"Tool-specific configurations" in the accompanying paper), without searching the
NT database.


## Evaluation context

For comparison, the RNAcentral and Rfam MSA sets released with OpenFold3 were
obtained. MSA depth is defined as the number of sequences in an alignment.
Effective sequence counts (N_eff) were computed with hhmake, and
length-normalized diversity as N_f = N_eff / sqrt(L), where L is the query
length. The paired comparison against the OpenFold3 RNAcentral set was
restricted to the 4,991 queries present in both collections.


## Contents

  Files              1,723 x batch_NNNN.tar.zst, plus aria2c_urls.txt
  Total size         1.957 TiB (2,151,286,583,636 bytes) compressed
  Batch numbering    batch_0000 through batch_1727
  Median batch size  1.53 GB (max 2.35 GB)
  Query sequences    1,731,677
  Alignment format   A3M, one file per query sequence

Note on numbering: five indices in the batch_0000 to batch_1727 range are
absent, namely 0363, 0729, 0730, 0731 and 0732. The numbering is therefore not
contiguous; 1,723 batch files exist in total. Scripts that iterate 0000..1727
blindly should tolerate missing indices.

The file aria2c_urls.txt lists the original upstream URLs and is retained for
provenance. It points at the source bucket, not at this mirror.


## Download

Anonymous, no credentials required.

Everything (1.957 TiB, check your disk first):

    aws s3 sync --no-sign-request \
      s3://steineggerlab/riboseek/rna_central_msa/ ./rna_central_msa/

A single batch:

    aws s3 cp --no-sign-request \
      s3://steineggerlab/riboseek/rna_central_msa/batch_0000.tar.zst .

With rclone, for better throughput on the full set:

    rclone copy :s3:steineggerlab/riboseek/rna_central_msa/ ./rna_central_msa/ \
      --s3-provider AWS --s3-region us-east-1 \
      --transfers 8 --multi-thread-streams 8 --s3-chunk-size 64M --progress

A manifest of all file URLs, suitable for aria2c -i or wget -i, is available at:

  https://steineggerlab.s3.us-east-1.amazonaws.com/riboseek/rna_central_msa/urls.txt


## Extracting

Each batch is a zstd-compressed tar archive.

Verify integrity without extracting:

    zstd -t batch_0000.tar.zst

Extract:

    tar --use-compress-program=unzstd -xf batch_0000.tar.zst


## License

Released under CC BY 4.0.  See the LICENSE file, or:

  https://creativecommons.org/licenses/by/4.0/

You may share and adapt this data, including commercially, provided you give
appropriate credit.

Upstream components carry their own terms: RNAcentral is released under CC0, and
EternaFold under its own license. Attribution for those remains with their
respective authors.


## Citation

If you use this dataset or Riboseek, please cite:

  Park S, Didi K, Favor A, Bushuiev A, Kim S, Mirdita M, Steinegger M.
  Fast remote nucleotide sequence alignment with Riboseek.
  bioRxiv (2026). doi:10.64898/2026.07.31.741718

    @article{park2026riboseek,
      title   = {Fast remote nucleotide sequence alignment with Riboseek},
      author  = {Park, Sukhwan and Didi, Kieran and Favor, Andrew and
                 Bushuiev, Anton and Kim, Soohyun and Mirdita, Milot and
                 Steinegger, Martin},
      journal = {bioRxiv},
      year    = {2026},
      doi     = {10.64898/2026.07.31.741718},
      url     = {https://www.biorxiv.org/content/10.64898/2026.07.31.741718v1}
    }


## Acknowledgment

We gratefully acknowledge the Jen-Hsun and Lori Huang Foundation for providing a
computational grant used for this study.