Datasets:

outpost-bio
/

Compass

dataset_info:
  - config_name: handuo
    features:
      - name: Sample ID
        dtype: large_string
      - name: Taxa
        list: string
      - name: Relative Abundances
        list: float64
      - name: SIC Name
        dtype: large_string
      - name: Control
        dtype: bool
      - name: ATC Class
        dtype: large_string
      - name: Split
        dtype: large_string
    splits:
      - name: train
        num_bytes: 6676068
        num_examples: 3168
      - name: validation
        num_bytes: 834508
        num_examples: 396
      - name: test
        num_bytes: 834508
        num_examples: 396
    download_size: 8165547
    dataset_size: 8345084
  - config_name: mastrorilli
    features:
      - name: Taxa
        list: string
      - name: Relative Abundances
        list: float64
      - name: Sample ID
        dtype: large_string
      - name: Degradation Rate
        dtype: float64
      - name: Drug
        dtype: large_string
      - name: Split
        dtype: large_string
    splits:
      - name: train
        num_bytes: 98959509
        num_examples: 9282
      - name: validation
        num_bytes: 32879888
        num_examples: 3084
      - name: test
        num_bytes: 32549383
        num_examples: 3053
    download_size: 163722477
    dataset_size: 164388780
  - config_name: mgnify-biomes
    features:
      - name: Taxa
        list: string
      - name: Relative Abundances
        list: float64
      - name: Run Accession
        dtype: large_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
      - name: Biome 1
        dtype: large_string
      - name: Biome 2
        dtype: large_string
      - name: Biome 3
        dtype: large_string
      - name: Biome 4
        dtype: large_string
      - name: Biome 5
        dtype: large_string
      - name: Split
        dtype: large_string
      - name: __index_level_0__
        dtype: int64
    splits:
      - name: train
        num_bytes: 343694740
        num_examples: 33121
      - name: validation
        num_bytes: 42950168
        num_examples: 4139
      - name: test
        num_bytes: 42950168
        num_examples: 4139
    download_size: 423262451
    dataset_size: 429595076
  - config_name: roswall
    features:
      - name: Sample ID
        dtype: large_string
      - name: Timepoint
        dtype: large_string
      - name: Delivery Mode
        dtype: large_string
      - name: Taxa
        list: string
      - name: Relative Abundances
        list: float64
      - name: Split
        dtype: large_string
    splits:
      - name: train
        num_bytes: 8135648
        num_examples: 1625
      - name: validation
        num_bytes: 1016330
        num_examples: 203
      - name: test
        num_bytes: 1016330
        num_examples: 203
    download_size: 10078148
    dataset_size: 10168308
configs:
  - config_name: handuo
    data_files:
      - split: train
        path: handuo/train-*
      - split: validation
        path: handuo/validation-*
      - split: test
        path: handuo/test-*
  - config_name: mastrorilli
    data_files:
      - split: train
        path: mastrorilli/train-*
      - split: validation
        path: mastrorilli/validation-*
      - split: test
        path: mastrorilli/test-*
  - config_name: mgnify-biomes
    data_files:
      - split: train
        path: mgnify-biomes/train-*
      - split: validation
        path: mgnify-biomes/validation-*
      - split: test
        path: mgnify-biomes/test-*
  - config_name: roswall
    data_files:
      - split: train
        path: roswall/train-*
      - split: validation
        path: roswall/validation-*
      - split: test
        path: roswall/test-*
license: apache-2.0

Compass

This dataset collects processed 16S / shotgun-derived microbiome tables formatted for machine learning. Each config is one benchmark task; column sets differ between configs, but within a config the train, validation, and test splits use the same schema.

Dataset summary

See our preprint for details

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

Config	Task theme	Typical use	Num examples
`mgnify-biomes`	MGnify sample → biome hierarchy	Biome prediction from genus-level abundances and study metadata	41399
`roswall`	Infant gut (PRJEB38986)	Timepoint and delivery mode prediction from gut profiles	2031
`handuo`	Species-in-context (drug exposure)	Classification with drug / ATC context and abundances	3960
`mastrorilli`	Drug degradation	Regression: degradation rate from community composition	15419

Each row includes a Split field (train, validation, or test) consistent with the Hugging Face split you load.

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

REPO_ID = "outpost-bio/Compass"  # adjust if your namespace/name differs

# One benchmark (DatasetDict: train, validation, test)
ds = load_dataset(REPO_ID, "mgnify-biomes")

train = ds["train"]
val = ds["validation"]
test = ds["test"]

# Optional: default config (library-defined ordering)
ds_default = load_dataset(REPO_ID)

Column reference

Taxa: Taxa and Relative Abundances are parallel lists containing taxon identifiers and relative abundances

`mgnify-biomes`

Context / labels: Run Accession, Data Type, Sequencing Method, Pipeline Version, Study Accession, Biome 1 … Biome 5 (biome lineage).
Features: Taxa, Relative Abundances
Split: Split

`roswall`

Sample / environment: Sample ID (run accession), Timepoint, Delivery Mode
Features: Taxa, Relative Abundances
Split: Split

`handuo`

Metadata: SIC Name, Control, ATC Class
Features: Taxa, Relative Abundances, Sample ID
Split: Split

`mastrorilli`

Targets / metadata: Degradation Rate, Drug, Sample ID
Features: Taxa, Relative Abundances
Split: Split
Splits follow sample boundaries (no overlap of Sample ID across splits).

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

Handuo Shi, Daniel P Newton, Taylor H Nguyen, Sylvie Estrela, Juan Sanchez, Michael Tu, Po-Yi Ho, Qinglin Zeng, Brian DeFelice, Justin Sonnenburg, and Kerwyn Casey Huang. Nutrient competition predicts gut microbiome restructuring under drug perturbations. August 2024.

Eleonora Mastrorilli, Pamela Herd, Federico E. Rey, Andrew L. Good- man, and Michael Zimmermann. Linking interpersonal differences in gut microbiota composition and drug biotransformation activity. bioRxiv, 2026. doi: 10.64898/2026.01.21.700809. URL https://www.biorxiv. org/content/early/2026/01/21/2026.01.21.700809.

Josefine Roswall, Lisa M Olsson, Petia Kovatcheva-Datchary, Staffan Nilsson, Valentina Tremaroli, Marie-Christine Simon, Pia Kiilerich, Rozita Akrami, Manuela Krämer, Mathias Uhlén, Anders Gummesson, Karsten Kristiansen, Jovanna Dahlgren, and Fredrik Bäckhed. Develop- mental trajectory of the healthy human gut microbiota during the first 5 years of life. Cell Host Microbe, 29(5):765–776.e3, May 2021.

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