cytetype 0.5.0

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

CyteType

---

CyteType is a Python package for deep chracterization of cell clusters from single-cell RNA-seq data. This package interfaces with Anndata objects to call CyteType API.

Quick Start

import anndata
import scanpy as sc
import cytetype

# Load and preprocess your data
adata = anndata.read_h5ad("path/to/your/data.h5ad")
sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)
sc.pp.pca(adata)
sc.pp.neighbors(adata)
sc.tl.leiden(adata, key_added = "clusters") 
sc.tl.rank_genes_groups(adata, groupby='clusters', method='t-test')

# Initialize CyteType (performs data preparation)
annotator = cytetype.CyteType(adata, group_key='clusters')

# Run annotation
adata = annotator.run(
    study_context="Human brain tissue from Alzheimer's disease patients"
)

# View results
print(adata.obs.cytetype_annotation_clusters)
print(adata.obs.cytetype_cellOntologyTerm_clusters)

Installation

pip install cytetype

Usage

Required Preprocessing

Your AnnData object must have:

• Log-normalized expression data in adata.X
• Cluster labels in adata.obs
• Differential expression results from sc.tl.rank_genes_groups

import scanpy as sc

# Standard preprocessing
sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)

# Clustering
sc.pp.pca(adata)
sc.pp.neighbors(adata)
sc.tl.leiden(adata, key_added='clusters')

# Differential expression (required)
sc.tl.rank_genes_groups(adata, groupby='clusters', method='t-test')

Annotation

from cytetype import CyteType

# Initialize (data preparation happens here)
annotator = CyteType(adata, group_key='clusters')

# Run annotation
adata = annotator.run(
    study_context="Adult human brain tissue samples from healthy controls and Alzheimer's disease patients, analyzed using 10X Genomics single-cell RNA-seq. Samples include cortical and hippocampal regions."
)

# Or with custom metadata for tracking
adata = annotator.run(
    study_context="Adult human brain tissue samples from healthy controls and Alzheimer's disease patients, analyzed using 10X Genomics single-cell RNA-seq. Samples include cortical and hippocampal regions.",
    metadata={
        'experiment_name': 'Brain_AD_Study',
        'run_label': 'initial_analysis'
    }
)

# Results are stored in:
# - adata.obs.cytetype_annotation_clusters (cell type annotations)
# - adata.obs.cytetype_cellOntologyTerm_clusters (cell ontology terms)
# - adata.uns['cytetype_results'] (full API response)

The study_context should include comprehensive biological information about your experimental setup:

• Organisms: Species being studied (e.g., "human", "mouse")
• Tissues: Tissue types and anatomical regions
• Diseases: Disease conditions or states
• Developmental stages: Age, developmental timepoints
• Single-cell methods: Sequencing platform (e.g., "10X Genomics", "Smart-seq2")
• Experimental conditions: Treatments, time courses, perturbations

Example: "Adult human brain tissue samples from healthy controls and Alzheimer's disease patients, analyzed using 10X Genomics single-cell RNA-seq. Samples include cortical and hippocampal regions."

Configuration Options

Initialization Parameters

annotator = CyteType(
    adata,
    group_key='leiden',                    # Required: cluster column name
    rank_key='rank_genes_groups',          # DE results key (default)
    gene_symbols_column='gene_symbols',    # Gene symbols column (default)
    n_top_genes=50,                        # Top marker genes per cluster
    aggregate_metadata=True,               # Aggregate metadata (default)
    min_percentage=10,                     # Min percentage for cluster context
    pcent_batch_size=2000,                 # Batch size for calculations
)

Submitting Annotation job

The run method accepts several configuration parameters to control the annotation process:

Custom LLM Configuration

The CyteType API provides access to some chosen LLM providers by default. Users can choose to provide their own LLM models and model providers. Many models can be provided simultaneously and then they will be used iteratively for each of the clusters.

adata = annotator.run(
    study_context="Human PBMC from COVID-19 patients",
    model_config=[{
        'provider': 'openai',
        'name': 'gpt-4o-mini',
        'apiKey': 'your-api-key',
        'baseUrl': 'https://api.openai.com/v1',  # Optional
        'modelSettings': {                       # Optional
            'temperature': 0.0,
            'max_tokens': 4096
        }  
    }],
)

Rate Limiting

If you do not provide your own model providers then the CyteType API implements rate limiting for fair usage:

• Annotation submissions: 5 requests per hour per IP
• Result retrieval: 20 requests per minute per IP

If you exceed rate limits, the system will return appropriate error messages with retry timing information

Supported providers: openai, anthropic, google, xai, groq, mistral, openrouter

Advanced parameters

adata = annotator.run(
    ...
    run_config={
        'concurrentClusters': 3,        # Default: 3, Range: 2-10
        'maxAnnotationRevisions': 2,    # Default: 2, Range: 1-5
        'maxLLMRequests': 500           # Default: 500, Range: 50-2000
    },
    
    # Custom metadata for tracking
    metadata={
        'experiment_name': 'PBMC_COVID_Study',
        'run_label': 'baseline_analysis',
        'researcher': 'Dr. Smith',
        'batch': 'batch_001'
    },
    
    # API polling and timeout settings
    poll_interval_seconds=10,           # How often to check for results
    timeout_seconds=1200,               # Max wait time (20 minutes)
    
    # API configuration
    api_url="https://custom-api.com",   # Custom API endpoint
    auth_token="your-auth-token",       # Authentication token
    save_query=True                     # Save query to query.json
)

Run configuration

• concurrentClusters (int, default=5, range=2-30): Maximum number of clusters to process simultaneously. Higher values may speed up processing but can cause rate limit errors from LLM API providers.
• maxAnnotationRevisions (int, default=2, range=1-5): Maximum number of refinement iterations based on reviewer feedback. More revisions may improve annotation quality but increase processing time.
• maxLLMRequests (int, default=500, range=50-2000): Maximum total number of LLM API calls allowed for the entire job. The job will be terminated if this limit is reached. Helps control costs and prevents runaway processes.

Additional Run Parameters

• metadata (dict, optional): Custom metadata to send with the API request for tracking purposes. Can include experiment names, run labels, researcher information, or any other user-defined data. This metadata is sent to the API but not stored locally with results.
• poll_interval_seconds (int, default=30): How frequently to check the API for job completion.
• timeout_seconds (int, default=3600): Maximum time to wait for results before timing out.
• api_url (str): Custom API endpoint URL for self-hosted deployments.
• auth_token (str, optional): Bearer token for API authentication.
• save_query (bool, default=True): Whether to save the API query to query.json for debugging.

Annotation Process

CyteType performs comprehensive cell type annotation through an automated pipeline:

Core Functionality

• Automated Annotation: Identifies likely cell types for each cluster based on marker genes
• Ontology Mapping: Maps identified cell types to Cell Ontology terms (e.g., CL_0000127)
• Review & Justification: Analyzes supporting/conflicting markers and assesses confidence
• Alternative Suggestions: Provides potential alternative annotations when applicable
• Real-time Progress: Updates results incrementally as clusters are processed

Job Status and Progress

When you submit an annotation job, it progresses through several stages:

• PENDING: Job is queued and waiting to start
• PROCESSING: Job is actively running with incremental results available
• COMPLETED: All clusters have been annotated successfully
• FAILED: Processing encountered an error

During PROCESSING, you can:

• Monitor progress through the report URL
• View partial results for completed clusters
• See biological context summaries
• Track completion status

Result Format

Results include detailed annotations for each cluster:

# Access results after annotation using the helper method
results = annotator.get_results()

# Or access directly from the stored JSON string
import json
results = json.loads(adata.uns['cytetype_results']['result'])

# Each annotation includes:
for annotation in results['annotations']:
    print(f"Cluster: {annotation['clusterId']}")
    print(f"Cell Type: {annotation['annotation']}")
    print(f"Confidence: {annotation['confidence']}")
    print(f"Ontology Term: {annotation['ontologyTerm']}")
    print(f"Supporting Markers: {annotation['supportingMarkers']}")
    print(f"Justification: {annotation['justification']}")

Example Report

View a sample annotation report: CyteType Report

Development

Setup

git clone https://github.com/NygenAnalytics/cytetype.git
cd cytetype
uv sync --all-extras
uv run pip install -e .

Testing

uv run pytest              # Run tests
uv run ruff check .        # Linting
uv run ruff format .       # Formatting
uv run mypy .              # Type checking

License

Licensed under CC BY-NC-SA 4.0 - see LICENSE for details.