CyteOnto: Semantic Cell Type Annotation Comparison Using Large Language Models and Cell Ontology

cyteonto compares two sets of cell type annotations against the Cell Ontology (CL). Given label lists from a study author and one or more annotation algorithms, it:

1. Decomposes mixture labels (doublets, mixed populations) into cell-type parts with an LLM when needed.
2. Generates a structured description for every label (or part) with an LLM.
3. Embeds those descriptions with a configured embedding model.
4. Matches each embedding to the closest CL term.
5. Scores each author/algorithm pair using an ontology-aware similarity metric (default: a Gaussian kernel on the cosine similarity of the CL term embeddings). Compound pairs default to the maximum entry in the part-by-part score matrix (compound_scoring="max"); set compound_scoring="hungarian_mean" for Hungarian assignment mean with a coverage penalty when part counts differ.
6. Returns a tidy DataFrame with one row per (algorithm, pair_index).

📚 Documentation: docs.nygen.io/CyteOnto covers the hosted API, the Python package, the similarity metrics, compound labels and caching.

Modal Service

The CyteOnto is available as a service! No API keys required, we provide it for you!

Quick start:

Send a compare request:

export CYTEONTO_URL="https://cyteonto.nygen.io"

curl -sS -X POST "$CYTEONTO_URL/compare" \
  -H 'Content-Type: application/json' \
  -d '{
    "authorLabels": ["alveolar macrophage", "regulatory T cell", "CD8-positive, alpha-beta T cell"],
    "algorithms": {
      "algo1": ["lung macrophage", "Treg", "CD8 T cell"],
      "algo2": ["alveolar mac", "T regulatory cell", "cytotoxic T cell"]
    }
  }'

Response:

{ "runId": "run-<uuid>", "state": "queued" }

Save the runId and poll status:

RUN_ID="run-<uuid>"
curl -sS "$CYTEONTO_URL/status/$RUN_ID" | jq

Fetch the result once state == "completed":

curl -sS "$CYTEONTO_URL/result/$RUN_ID?format=json" | jq
curl -sS "$CYTEONTO_URL/result/$RUN_ID?format=csv" -o "$RUN_ID.csv"

The hosted API quickstart walks through these steps, and the HTTP API reference lists every request field. Check out the modal_app/example.py file to run via Python client.

Python package

cyteonto is not published on PyPI. Install it from GitHub and download the Cell Ontology files once:

pip install "git+https://github.com/NygenAnalytics/CyteOnto.git"
python -m cyteonto.setup

The Python quickstart compares labels with your own LLM and embedding API keys.

Citation

CyteOnto is a part of the CyteType. If you use CyteOnto in your research, please cite our preprint:

Ahuja G, Antill A, Su Y, Dall'Olio GM, Basnayake S, Karlsson G, Dhapola P. Multi-agent AI enables evidence-based cell annotation in single-cell transcriptomics. bioRxiv 2025. doi: 10.1101/2025.11.06.686964

@article{cytetype2025,
  title={Multi-agent {AI} enables evidence-based cell annotation in single-cell transcriptomics},
  author={Ahuja, Gautam and Antill, Alex and Su, Yi and Dall'Olio, Giovanni Marco and Basnayake, Sukhitha and Karlsson, G{\"o}ran and Dhapola, Parashar},
  journal={bioRxiv},
  year={2025},
  doi={10.1101/2025.11.06.686964},
  url={https://www.biorxiv.org/content/10.1101/2025.11.06.686964v1}
}