tysserand

Fast and accurate spatial networks reconstruction.

tysserand is a Python library to reconstruct spatial networks from spatially resolved omics experiments. It is intended as a common tool where the bioinformatics community can add new methods to reconstruct networks, choose appropriate parameters, clean resulting networks and pipe data to other libraries.
You can find the article and supplementary information on Bioinformatics, and the freely available preprint (same text!) is on BioRxiv.
A turorial is available here

tysserand is fast: it is 50 to more than 120 times faster than PySAL.
tysserand is accurate: it implements the best performing methods, tested on simulated and real bioimages. tysserand is user friendly and interactive: it integrates tools to choose appropriate parameters and facilitates the use of napari-based interactive image visualization and network annotation. tysserand is modular and opened to contributions: if you have an idea on how to improve reconstruction methods, create a particular one for a specific case, or make them even faster, join us!

Installation

Simply do

pip install tysserand

If you want the latest features not published on PyPI run

pip install git+https://github.com/VeraPancaldiLab/tysserand.git

It is best practice to create a dedicated environment for each project. To do it with pyenv:

# create environment
pyenv install 3.10.13
pyenv virtualenv 3.10.13 spatial-networks
# add environment to jupyterlab
pyenv activate spatial-networks
ipython kernel install --user --name=spatial-networks

or with Conda and Mamba:

# create environment
conda install mamba -n base -c conda-forge
mamba env create --file environment.yml
# add environment to jupyterlab
conda activate spatial-networks
ipython kernel install --user --name=spatial-networks

If you want to reproduce results in the publication, install also PySAL

pip install libpysal geopandas fiona shapely pyproj rtree

Implemented methods

Delaunay triangulation

This method builds virtual cells centered arround each node and contacting each other to fully tile the space occupyied by the nodes. Edges are drawn between the nodes of contacting tiles. The node_adaptive_trimming option can infer a better distance threshold to discard edges for each individual node.

k-nearest neighbors

Each node is linked with its k nearest neighbors. It is the most common method used in single cell publications, althought it produces artifacts well visible on simple 2D networks.

radial distance neighbors

Each node is linked to nodes closer than a threshold distance D, that is to say each node is linked to all nodes in a circle of radius D.

Area contact

Nodes are the center of detected objects (like after cell segmentation) and they are linked if their respective areas are in contact or closer than a given distance threshold.
A parallelized version is implemented with the Dask library.

Area contact and k-nearest neighbors

Areas are linked if they are in contact or closer than a given distance. Then, the remaining non connected areas are connected with their nearest neighbors.