Datasets:

outpost-bio
/

Atlas

dataset_info:
  features:
    - name: Taxa
      list: string
    - name: Relative Abundances
      list: float64
    - name: Run Accession
      dtype: string
    - name: Data Type
      dtype: large_string
    - name: Sequencing Method
      dtype: large_string
    - name: Pipeline Version
      dtype: large_string
    - name: Study Accession
      dtype: large_string
  splits:
    - name: pretrain
      num_bytes: 5027475729
      num_examples: 485377
    - name: benchmark
      num_bytes: 557448072
      num_examples: 53931
  download_size: 5552785595
  dataset_size: 5584923801
configs:
  - config_name: default
    data_files:
      - split: pretrain
        path: data/pretrain-*
      - split: benchmark
        path: data/benchmark-*
license: apache-2.0

Atlas

Dataset summary

See our preprint for details

Join our slack community for support and discussion about microbiome foundation models.

Each row is one sample. Taxa and Relative Abundances are aligned lists (same length per row): semicolon-separated rank strings for taxa, and matching relative abundances. Additional columns record data type, sequencing method, pipeline version, and study accession.

Data come from processed abundance matrices, filtered, then randomly split into pretrain (90%) and benchmark (10%) with seed=42.

Splits

Split	Description	Num Examples
`pretrain`	Pretraining subset.	485377
`benchmark`	Held-out benchmark subset.	53931

Data fields

Column	Type	Description
Taxa	list of string	Taxon labels.
Relative Abundances	list of float	Relative abundances in the same order as Taxa.
Sample	string	Sample identifier, if present.
Data Type, Sequencing Method, Pipeline Version, Study Accession	string	Provenance metadata; exact set of columns matches the published Parquet schema.

Filtering

Filtering used:

min_relative_abundance: 0.0001
min_num_taxa: 10

Usage

This repository is gated. To use it you'll need to:

Request access — click the "Request access" button at the top of this repo's page on Hugging Face. Requests are auto-approved.
Authenticate — log in to Hugging Face from your environment so the download tooling can use your token:

   huggingface-cli login

Or set the token directly:

   export HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxx

You can create a token at https://huggingface.co/settings/tokens.

Once both steps are done, you can load the model/dataset normally:

from datasets import load_dataset

ds = load_dataset("outpost-bio/Atlas")
row = ds["pretrain"][0]
taxa = row["Taxa"]
abund = row["Relative Abundances"]

License

apache-2.0

References to public data used

Lorna Richardson, Ben Allen, Germana Baldi, Martin Beracochea, Maxwell L Bileschi, Tony Burdett, Josephine Burgin, Juan Caballero- Pérez, Guy Cochrane, Lucy J Colwell, Tom Curtis, Alejandra Escobar- Zepeda, Tatiana A Gurbich, Varsha Kale, Anton Korobeynikov, Shriya Raj, Alexander B Rogers, Ekaterina Sakharova, Santiago Sanchez, Dar- ren J Wilkinson, and Robert D Finn. Mgnify: the microbiome sequence data analysis resource in 2023. Nucleic Acids Research, 51(D1):D753– D759, 12 2022. ISSN 0305-1048. doi: 10.1093/nar/gkac1080. URL https://doi.org/10.1093/nar/gkac1080

Citation

Learning the Language of the Microbiome with Transformers
Neythen J Treloar, Saif Ur-Rehman, Jenny Yang
bioRxiv 2026.05.02.722381; doi: https://doi.org/10.64898/2026.05.02.722381