Datasets:

neuralbioinfo
/

eskapee

license: cc-by-nc-nd-4.0
dataset_info:
  features:
    - name: assembly
      dtype: large_string
    - name: sequence_id
      dtype: int64
    - name: taxon
      dtype: int64
    - name: taxon_name
      dtype: large_string
    - name: taxon_short
      dtype: large_string
    - name: contig
      dtype: large_string
    - name: seq_len
      dtype: int64
    - name: description
      dtype: large_string
    - name: strand
      dtype: large_string
    - name: sequence
      dtype: large_string
  splits:
    - name: train
      num_bytes: 26312382
      num_examples: 14
  download_size: 12213001
  dataset_size: 26312382
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

pretty_name: ESKAPE contigs (ProkBERT-ready) tags: - genomics - dna - bacteria - eskape - transformers - prokbert - contrastive-learning task_categories: - feature-extraction - text-classification license: other language: - en

ESKAPE contigs (ProkBERT-ready)

Dataset summary

This dataset contains nucleotide contigs/genomes for ESKAPE pathogens, packaged for transformer-based modeling (e.g., ProkBERT). It includes NCBI taxonomy identifiers and compact scientific labels that are convenient for visualization.

Data structure

Rows

Each row corresponds to one contig (chromosome or plasmid) from an assembly.

Columns

assembly (str): Assembly accession (e.g., GCF_...).
taxon (int): NCBI Taxonomy ID.
taxon_name (str): Parsed scientific name (Genus species [+ subsp.]).
taxon_short (str): Short visualization label (e.g., E. coli MG1655).
contig (str): Contig accession/identifier.
description (str): NCBI contig description.
seq (str): Nucleotide sequence (uppercase).
seq_len (int): Sequence length.

Intended use

Pretraining / representation learning on bacterial genomes/contigs.
Supervised or weakly-supervised tasks with taxon-level labels.
Benchmarking sequence encoders and contrastive objectives.

Notes / caveats

taxon_name and taxon_short are derived heuristically from the contig description to standardize labels for plots. If you need strict taxonomy-name mapping, resolve taxon via an authoritative taxonomy dump/API.
Verify licensing/redistribution constraints for your specific upstream sources before downstream redistribution.

Usage

from datasets import load_dataset

ds = load_dataset("neuralbioinfo/eskapee")
ds["train"][0]

Contact Information

For any questions, feedback, or contributions regarding the datasets or ProkBERT, please feel free to reach out:

Name: Balázs Ligeti
Email: obalasz@gmail.com

We welcome your input and collaboration to improve our resources and research.

Citation

@Article{ProkBERT2024,
  author  = {Ligeti, Balázs et al.},
  journal = {Frontiers in Microbiology},
  title   = {{ProkBERT} family: genomic language models},
  year    = {2024},
  volume  = {14},
  URL     = {https://www.frontiersin.org/articles/10.3389/fmicb.2023.1331233},
  DOI     = {10.3389/fmicb.2023.1331233}
}