cytetype 0.19.3

Python client for characterization of clusters from single-cell RNA-seq data.

CyteType

Agentic, Evidence-Based Cell Type Annotation for Single-Cell RNA-seq

CyteType performs automated cell type annotation in single-cell RNA sequencing (scRNA-seq) data. It uses a multi-agent AI architecture to deliver transparent, evidence-based annotations with Cell Ontology mapping.

Integrates with Scanpy and Seurat workflows.

---

Preprint published: Nov. 7, 2025: bioRxiv link - Dive into benchmarking results

---

Why CyteType?

Cell type annotation is one of the most time-consuming steps in single-cell analysis. It typically requires weeks of expert curation, and the results often vary between annotators. When annotations do get done, the reasoning is rarely documented; this makes it difficult to reproduce or audit later.

CyteType addresses this with a novel agentic architecture: specialized AI agents collaborate on marker gene analysis, literature evidence retrieval, and ontology mapping. The result is consistent, reproducible annotations with a full evidence trail for every decision.

---

Key Features

Feature	Description
Cell Ontology Integration	Automatic CL ID assignment for standardized terminology and cross-study comparison
Confidence Scores	Numeric certainty values (0–1) for cell type, subtype, and activation state — useful for flagging ambiguous clusters
Linked Literature	Each annotation includes supporting publications and condition-specific references — see exactly why a call was made
Annotation QC via Match Scores	Compare CyteType results against your existing annotations to quickly identify discrepancies and validate previous work
Embedded Chat Interface	Explore results interactively; chat is connected to your expression data for on-the-fly queries

Also included: interactive HTML reports, Scanpy/Seurat compatibility (R wrapper via CyteTypeR), and no API keys required out of the box.

📹 Watch CyteType intro video

---

Quick Start

Installation

pip install cytetype

Basic Usage with Scanpy

import scanpy as sc
from cytetype import CyteType

# Assumes preprocessed AnnData with clusters and marker genes
group_key = 'clusters'
annotator = CyteType(
    adata, 
    group_key=group_key, 
    rank_key='rank_genes_' + group_key, 
    n_top_genes=100
)
adata = annotator.run(study_context="Human PBMC from healthy donor")
sc.pl.umap(adata, color='cytetype_annotation_clusters')

🚀 Try it in Google Colab

Note: No API keys required for default configuration. See Configuration for LLM setup, artifact handling, and advanced options.

Using R/Seurat? → CyteTypeR

---

Documentation

Resource	Description
Configuration	LLM settings, parameters, and customization
Output Columns	Understanding annotation results and metadata
Troubleshooting	Common issues and solutions
Development	Contributing and local setup
Discord	Community support

---

Output Reports

Each analysis generates an HTML report documenting annotation decisions, reviewer comments and an embedded chat interface for further exploration.

View example report

---

Benchmarks

Validated across PBMC, bone marrow, tumor microenvironment, and cross-species datasets. CyteType's agentic architecture consistently outperforms existing annotation methods:

Comparison	Improvement
vs GPTCellType	+388%
vs CellTypist	+268%
vs SingleR	+101%

Browse CyteType results on atlas scale datasets

---

Citation

If you use CyteType 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={Gautam Ahuja, Alex Antill, Yi Su, Giovanni Marco Dall'Olio, Sukhitha Basnayake, Göran Karlsson, Parashar Dhapola},
  journal={bioRxiv},
  year={2025},
  doi={10.1101/2025.11.06.686964},
  url={https://www.biorxiv.org/content/10.1101/2025.11.06.686964v1}
}

---

License

CyteType is free for academic and non-commercial research under CC BY-NC-SA 4.0.

For commercial licensing, contact [e-mail].

---