MultiLLMc

Multi-LLM Collaborative Cell Type Annotation for Single-Cell RNA-seq

A framework that uses multiple large language models to collaboratively annotate cell types in single-cell RNA sequencing data.

Features

• Multi-LLM Discussion: Leverage multiple LLMs to discuss and annotate cell types
• Quality Review: Built-in review mechanism to audit and correct annotations
• Consensus Building: Harmonize results from multiple models to reach consensus
• Flexible Provider Support: Support for various API providers (n1n, openrouter)

Installation

pip install multillmc

Or install from source:

git clone https://github.com/51cat/multiLLMc.git
cd multiLLMc
pip install -e .

Quick Start

from mllmcelltype import Seminar, Reviewer, Harmonizer

# Define marker genes for each cluster
marker_genes = {
    "cluster_0": ["CD3D", "CD3E", "CD3G", "IL7R", "TCF7"],
    "cluster_1": ["CD79A", "CD79B", "MS4A1", "IGHM", "IGKC"],
    "cluster_2": ["CD14", "LYZ", "S100A8", "S100A9", "FCN1"],
}

# Initialize seminar
seminar = Seminar(
    marker_dict=marker_genes,
    species='human',
    tissue='PBMC'
)

# Configure API and models
seminar.set_api("your-api-key")
seminar.set_model_list(['gpt-4o', 'claude-3-sonnet', 'gemini-pro'])
seminar.set_provider('openrouter')

# Run annotation
seminar.make_init_ann_promopt('major_celltype')
seminar.start()

# Get results
cluster_results = seminar.get_cluster_results()

# Optional: Review and harmonize
reviewer = Reviewer(seminar)
reviewer.set_api("your-api-key")
reviewer.set_provider('openrouter')
reviewer.get_seminar_results()
reviewer.review()

harmonizer = Harmonizer(seminar)
harmonizer.set_api("your-api-key")
harmonizer.set_provider('openrouter')
harmonizer.get_seminar_results()
harmonizer.check()
consensus = harmonizer.get_check_result()

Workflow

1. Seminar: Multiple LLMs independently annotate cell types based on marker genes
2. Review: Audit annotations for quality, detect hallucinations, and correct errors
3. Harmonize: Build consensus from multiple model predictions

API Providers

Supported providers:

• n1n: N1N API Gateway
• openrouter: OpenRouter unified API

Requirements

• Python >= 3.10
• langchain >= 0.2.0
• langchain-openai >= 0.1.0
• pydantic >= 2.0.0
• jinja2 >= 3.1.0

License

MIT License

Author

51cat