Single-Cell Atlas Builder is an open-source platform for building, analyzing, and visualizing single-cell RNA-seq atlases.
It integrates widely used single-cell analysis tools (Scanpy, CellTypist, pySCENIC) with a FastAPI backend, SQLAlchemy-managed database, and an optional Streamlit frontend for interactive visualization. LLaMA 3 is used for cluster/pathway summarization.
┌────────────────────────┐
│ Client UI │
│ (Streamlit / Dash / │
│ Swagger) │
└────────────┬──────────┘
│ REST API
┌────────────▼──────────┐
│ FastAPI Server │
│ /upload /integrate / │
│ /visualize │
└────────────┬──────────┘
│ ORM
┌────────────▼──────────┐
│ SQLAlchemy ORM │
│ PostgreSQL / SQLite │
└────────────┬──────────┘
│ Data
┌────────────▼──────────┐
│ Analysis Engine │
│ Scanpy, CellTypist, │
│ LLaMA 3, pySCENIC │
└───────────────────────┘
- Upload & Manage Datasets: Supports
.h5ad,.csv,.mtxformats. - Preprocessing: QC, filtering, normalization, PCA, UMAP.
- Clustering & Annotation: Leiden clustering and optional CellTypist integration.
- Integration: Merge multiple datasets into a unified atlas.
- Interactive Visualization: Streamlit-based UMAP and gene expression plots.
- LLM Summarization: Optional cluster/pathway summaries using LLaMA 3.
- Reproducible Deployment: Docker-ready for local or cloud deployment.
Streamlit UMAP visualization example
scatlas-builder/
├── app/
│ ├── main.py # FastAPI entrypoint
│ ├── routers/ # API endpoints: upload, integrate, visualize
│ ├── services/ # Processing and clustering functions
│ ├── schemas.py # Pydantic models
│ ├── tables.py # SQLAlchemy ORM models
│ ├── database.py # Database connection
│ ├── utils/ # File and helper utilities
│ └── ui/ # Streamlit application
├── uploads/ # Uploaded datasets
├── docker/ # Dockerfile
├── requirements.txt
└── README.md
- Python 3.9+
- Conda recommended for scientific packages
- SQLite (default) or PostgreSQL
git clone https://github.com/yourusername/scatlas-builder.git
cd scatlas-builder
pip install -r requirements.txtuvicorn app.main:app --reload- API docs:
http://127.0.0.1:8000/docs - Streamlit UI:
streamlit run app/ui/streamlit_app.py
docker build -t scatlas-builder .
docker run -p 8000:8000 scatlas-builder| Endpoint | Method | Description |
|---|---|---|
/upload |
POST | Upload a dataset with metadata |
/integrate/run/{dataset_id} |
POST | Run preprocessing, clustering, and integration pipeline |
/visualize/umap/{dataset_id} |
GET | Generate UMAP plot of processed dataset |
from app.services import preprocessing, clustering, llm_summary
import scanpy as sc
# Load dataset
adata = preprocessing.load_input("uploads/sample.h5ad")
# Preprocessing
adata = preprocessing.run_qc_and_normalize(adata)
adata = preprocessing.run_pca_umap(adata)
# Clustering
adata = clustering.run_leiden(adata)
# Plot UMAP
sc.pl.umap(adata, color='leiden')
# Optional LLaMA 3 summarization
markers = {'0': ['GeneA', 'GeneB'], '1': ['GeneC', 'GeneD']}
summary = llm_summary.summarize_cluster_markers(markers)
print(summary)- Add CellTypist integration for automatic cell-type annotation
- LLaMA 3-based cluster/pathway summarization
- Celery + Redis for background processing of large datasets
- Streamlit dashboard enhancements for interactive exploration
- Multi-dataset integration (Harmony, scVI)
If you use this project, please cite:
Manish Kumar, Single-Cell Atlas Builder, 2025.
GitHub: https://github.com/man4ish/scatlas-builder