This repository provides data and code to reproduce data presented in "Spatial Transcriptomics to define transcriptional patterns of zonation and structural components in the liver", and show-cases the applicability and potential to study tissue transcriptomics computationally. In this study, Spatial Transcriptomics was performed on liver tissue sections of female wildtype mice of 8-12 weeks of age. All scripts presented here are designed to be used with any kind of spatial transcriptomics data and provide detailed documentation for the reproduction of the data presented in this original study. Original data and large files are placed at an external site save resources inlcuding count matrices, spot files, HE-images, masks, h5ad-files, etc. and can be accessed at zenodo.

• Overview
• System Requirements
• hepaquery
  • Dependencies
  • Installation
  • Data Preparation
• Data Access
• License

Below is an overview of the structure of this repository, including brief descriptions of respective item.

• data - contains processed data to be used in the analysis (files must be downloaded from an external resource, see below)

  • gene lists
    • marker genes - contains csv files with central and portal markers of the ST data and the SC data analyis performed for the liver data of the Mouse Cell Atlas
    • stereoscope - contains lists of 5000 most variable genes of single cell Mouse Cell Atlas data and Halpern et al study for sterescope analysis
    • veins - contains shortlists of marker genes for portal and central veins for expression by distance analysis
  • sterescope/sc - single cell data used for the stereoscope analysis

• scripts - contains processing scripts and notebooks

  • Liver-ST.Rmd - contains a R markdown script to perform canonical correlation analyis, clustering and DGEA, tissue visualization, correlation analysis, visualization of single cell integration using single cell data of the Mouse Cell Atlas and comparative analyses with published data from Halpern et al
  • MultiCCA.R - contains code for the modified canonical correlation analysis function used in Liver-ST.Rmd
  • cluster-interaction-analysis.ipynb - notebook outlining the cluster interaction analysis (used to produce Supplementary Image 2)
  • synthetic-data-analysis.ipynb - generation and analysis of synthetic spatial transcriptomics data, to compare correlation values between observed and random cell distributions.
  • make-gene-list.R - script to generate list of highly variable genes (hvgs) to use in stereoscope mapping.
  • prepare-data.py - program with CLI to generate h5ad files for spatial analysis (in vein-analysis.ipynb). See
  • vein-analysis.ipynb - notebook outlining the feature by distance analysis and vein type classification/prediction based on NEPs.
  • proportion-analysis.ipynb - similar to first part of vein-analysis.ipynb but looking at the cell type proportions -compared to expression levels - (obtained from stereoscope) as a function of distance to the nearest vein.

• res/sterescope-res/

  • CNX_ZY - folder for each sample (X,Y, and Z are various parts of identifiers). Each folder contains a W*.tsv file which are the sterescope results, formatted as [n_spots]x[n_types] matrices.
  • st_loss.txt - loss output for st-model
  • sc_loss.txt - loss output for sc-model
  • st_model.pt.gz - gzipped fitted st-model
  • sc_model.pt.gz - gzipped fitted sc-model
  • R.tsv.gz - gzipped estimated R paramters (in stereoscope model)
  • logits.tsv.gz - gzipped esitmated logits parameters (in stereoscope model)

• hepaquery - files constituting the hepaquery package

• setup.py - installation file for hepaquery module

The code should be compatible with all systems that support R and python. We recommend the following software versions:

• R >= 3.5
• python >= 3.7

To reproduce the analysis presented in the jupyter-notebooks and R markdown files, the packages listed below must be installed

• jupyter-notebook | LINK | Python
• rmarkdown | LINK | R

To facilitate reproduction of our results and enable easy exploration of similar data sets using the methods we present in this work, we have packaged these into a Python module called hepaquery. This, among other things, contains functions to generate feature_by_distance plots, classification of vein types based on neighborhood expression profiles (NEPs), and evaluation of prediction results.

The hepaquery package mainly relies on the python scientific framework, and will be installed automatically when running the commands in the next section. However, to list the packages and their recommended versions:

scikit-misc
numpy>=1.19.0
pandas>=1.0.0
anndata>=0.7.5
scipy>=1.5.4
scikit-learn>=0.23.2
matplotlib>=3.3.3

To install this package, do:

1. Enter a terminal window, change directory to this repository and type

$> python3 setup.py install

Depending on your OS and user configurations, you might have to add --user for this to work. The installation takes only a few seconds on a standard laptop computer.

2. Next, to test if the installation work, enter the following into the terminal:

$> python3 -c "import hepaquery; print(hepaquery.__version__)"

If everything went as expected, this should print the version of hepaquery in your terminal.

The notebook scripts/vein-analysis.ipynb illustrates the usage of hepaquery. Here the examples of the expected output when running hepaquery is found, and instructions regarding how to reproduce the plots presented in the manuscript. The analysis on the data in this study takes less than 5 minutes when run on a laptop computer.

To conduct the feature by distance analysis, the data must first be curated and formatted. For this purpose we provide the scripts/prepare-data.py script, which offers a convenient CLI to easily produce the required h5ad files from the raw count matrices, spot-files and HE-images. One additional key element are the masks indicating veins in the tissue. These masks should align with the HE-images (same dimensions), and have each veins colored by their class. YAML files are then used to specify which elements to include in the assembly of the h5ad object.

For each item associated with a unique sample, create a YAML file (filename.yaml) with the following structure:

count_data: PATH_TO_COUNT_FILE
spot_data: PATH_TO_SPOT_FILE
image: PATH_TO_HE-IMAGE_FILE
mask: PATH_TO_MASK_FILE
rgb:
  class1: [R1,G1,B1]
  class2: [R2,G2,B2]
  class3: [R3,G3,B3]

where class1 indicate the name of the first class (e.g., central) and [R1,G1,B1] gives the RGB value that indicate class1 in the mask. This is what we refer to as a configuration file

Next, simply go to the scripts folder and run in a terminal run:

$> python3 ./prepare-data.py -i CONFIG_FILE.yaml -n N_TYPES -s -d -o OUT_DIR

And h5ad files containing the essential information used in the vein-analysis.ipynb files will be produced. For more information regarding the parameters that may be used, simply do python3 ./prepare-data.py -h.

While GitHub supports storage of large files via the LFS system, we have placed our files at an external site to prevent unnecessary use of resources. The count matrices, spot files, HE-images and masks can be accessed at this link. To download all data and place it in the expected (by the scripts) location, you can also go to scripts, open a terminal and do:

$> chmod +x ./fetch-data.sh
$> ./fetch-data.sh ZENDO_LINK

In case the above does not work for you, simply go to the link referenced above, download the files and place the content of Hepaquery_data.zip in the data folder. This should be equivalent to what the script is doing for you.

This work is covered under the MIT License.