Pantheon_llm - Single-Cell Language Model Analysis Framework

A unified framework for single-cell RNA-seq analysis using large language models including scGPT, scFoundation, Geneformer, CellPLM, and UCE.

Table of Contents

• Installation
• Quick Start
• Supported Models
• Core Usage

Installation

pip install torch scanpy anndata
# Install model-specific dependencies as needed

Quick Start

import scanpy as sc
from model_factory import SCLLMManager

# Load your data
adata = sc.read_h5ad('your_data.h5ad')

# Initialize manager with a model
manager = SCLLMManager(
    model_type="scgpt",  # or "scfoundation", "geneformer", "cellplm", "uce"
    model_path="path/to/model",
    device="cuda"  # or "cpu", "auto"
)

# Get cell embeddings
embeddings = manager.get_embeddings(adata)
adata.obsm['X_scgpt'] = embeddings

Supported Models

Model	Purpose	Key Features
scGPT	General-purpose	Cell annotation, integration, generation
scFoundation	Foundation model	Large-scale pre-training, versatile
Geneformer	Gene perturbation	In silico perturbation experiments
CellPLM	Pre-training of Cell Language Model	cell-cell relations
UCE	Universal embeddings	Cross-species, cross-platform

Core Usage

You can find the detailed tutorial in the folder tutorials

Development

You can also read our DEVELOPER_GUIDE.md to know the model architecture

License

This project is licensed under the MIT License.