Pyxidis

Here we present Pyxidis, an integrated pipeline to build a graph-based representation of large 2D and 3D tissue microscopy images. It incorporates a tool to allows segmentation of large images (cutting them in small tiles for parallel computation and reconstruction of the image), a deep-learning based approach for cell classification as well as a graph-based analysis of spatial structures. An interactive selection tool is also provided to manually anotate regions. Pyxidis is described and used in the article "Spatiotemporal dynamics of cytokines expression dictate fetal liver hematopoiesis". This analysis pipeline and type of data analysis can be extended to large variety of images of biological tissues.

This repository contains a step by step tutorial to use this pipeline as well as example data to run main operations. Example images can be dowloaded from our Zenodo repository DOI:10.5281/zenodo.7867025.

Installation

We recommand to create a specific python environment to avoid any conflict in package versions.

conda create --name FL_image_analysis python=3.8

You need to install via the repository. In the command prompt enter:

git clone https://github.com/BaroudLab/Pyxidis.git

This should download the project. Then install the required packages. In the terminal, navigate to the package directory and type:

pip install .

The library is now installed on your computer.

For some applications, you will need to install a napari plugin for griottes. You can do it directly in the interface of napari (Plugin > Install packages > napari-griottes). More documentation is provided here: https://github.com/BaroudLab/napari-griottes. A part of the analysis presented here rely on the use of Griottes tool. More documentation is provided here: https://github.com/BaroudLab/Griottes and in the corresponding article: Griottes: a generalist tool for network generation from segmented tissue images.

Description of repository content

This repository contains all steps to reproduce image analysis pipeline (More description can be found in Fig2 and Methods of the article). Data files to run notebooks on examples are provided directly in the Github repository. Images can be downloaded from Zenodo (DOI:10.5281/zenodo.7867025).

The repository contains:

• Notebooks:

  • Image_processing: step by step tutorial to Pyxidis as well as the main type of data analysis. csv and txt files are provided in Data to run notebooks 4 and 5 on an example and corresponding images are on Zenodo.
  • Visualization: To evaluate good accuracy of segmentation and classification, it is convinient to have a graphical tool to overlap images and classification results. For this we use napari interface. Here, you can find notebooks to do it on 2D and 3D examples. csv files are also provided for these notebooks in a folder called Data and images on Zenodo.

• src: useful packages that are automatically installed in the installation step:

  • plot_data: some functions to allow easy visualization of classified data as dot plots.
  • coloriage: a graphical tool to manually select cells in a graph representation of the tissue.
  • saucisson: a set of functions to cut a big image in small pieces to allow parallel segmentation followed by reconstruction of the entire segmented image.

Image analysis steps (Image_processing)

• Training Neural Network for cell classification based on membrane staining: 0-Training_network.ipynb.
• Cut images in pieces for parallel segmentation using CellPose on GPUs / rebuild segmented image: 1-Saucisson_preprocessing.ipynb.
• Classify cells using Neural Network: 2-Classifying_cells.ipynb.

• Threshold for nuclear staining and gather all data from classification steps: 3-Generating_final_data_file.ipynb.
• Some typical spatial analysis: 4-Example_image_analysis.ipynb.
• Graph representation of the tissue and analysis of cells neighborhood: 5-Contact_on_graph_vs_random.ipynb.

Description of provided data

Data for visualization (in Notebooks>Visualization>Data)

Contain .csv files to overlayed segmentation and classification over the original image with napari on a 2D and 3D images for visual inspection

• data_example_2D.csv
• data_example_3D.csv

Data for image processing notebooks (in Notebooks>Image_processing>Data)

• 221212_6_labeled.csv: cell positions and average intensity on different channels in nucleus masks.
• CD45_NN_classification.csv: result of neural network classification of CD45.
• Kit_NN_classification: result of neural network classification of KIT.
• 221212_6_classified.csv: result of cell classification.
• 221212_6_classified_corrected.csv: result of cell classification after removing autofluorescent cells and disrupted regions.
• border_selection.txt: manual selection of border cells with Coloriage.
• vessels_selection.txt: manual selection of vessel with Coloriage.
• autofluorescent_cells.txt: autofluorescent cells to remove detected using neural network.
• manual_selection_cells_to_remove.txt: disrupted regions manually selected with Coloriage.

Data on Zenodo DOI:10.5281/zenodo.7867025

• 221212_6.tif: original image (containing 2 membrane stainings (CD45 and KIT) and a nuclear staining (LHX2)
• 221212_6_labeled.tif: image with segmentation after reconstruction.
• 221212_6_napari.tif: file to drag and drop in napari (different channel order).
• example_image_2D_napari.tif: tif files to overlay with classification to go with data_example_2D.csv
• example_image_3D_napari.tif: tif files to overlay with classification to go with data_example_3D.csv