Gaussian Process Spatial Alignment (GPSA)

The gpsa-engelhardt package implements Gaussian Process Spatial Alignment, a probabilistic model for aligning spatial genomics data into a shared coordinate system using deep Gaussian processes.

Install name: gpsa-engelhardt
Import name: gpsa

Paper: Alignment of spatial genomics and histology data using deep Gaussian processes
➤️ https://www.biorxiv.org/content/10.1101/2022.01.10.475692v1

View the Code on GitHub

🚀 Installation

pip install gpsa-engelhardt

# Usage
import gpsa
from gpsa.models import GPSA, VariationalGPSA

Requires Python 3.10+ and PyTorch.

🔬 Overview

gpsa provides two primary classes:

• GPSA — core generative model for probabilistic spatial alignment
• VariationalGPSA — variational approximation for scalable inference

Use GPSA to jointly model multiple spatial genomics datasets and correct spatial misalignments across experiments or modalities.

🧪 Example (Test the published PyPI package)

A minimal, runnable example is provided in examples/grid_example.py. It simulates a small synthetic dataset and runs GPSA alignment.

# Make a new virtual environment (Python 3.11 shown; 3.10 also works)
python3.11 -m venv gpsa_test_venv

# Activate the virtual environment
source gpsa_test_venv/bin/activate

# (optional) Upgrade pip
pip install --upgrade pip

# Clone the repository (for the example script)
git clone https://github.com/engelhardtgpsa/spatial-alignment.git
cd spatial-alignment

# Install GPSA from PyPI (pin to a specific version if desired)
pip install gpsa-engelhardt==0.6.15

# Run the example
python examples/grid_example.py

# Deactivate the virtual environment when done
deactivate

📊 Visualization

Example output showing the alignment of two misaligned spatial views:

The aligned coordinates converge during training:

Note that GUI backends (e.g., matplotlib with tkinter) may require extra setup on some systems.

🐞 Bug Reports

Please open issues at:
https://github.com/engelhardtgpsa/spatial-alignment/issues

📔 Citation

If you use GPSA in your work, please cite:

Jones, A. C., et al. Alignment of spatial genomics and histology data using deep Gaussian processes. bioRxiv (2022).
https://www.biorxiv.org/content/10.1101/2022.01.10.475692v1

📜 License

Apache-2.0