Skip to content

Latest commit

 

History

History
131 lines (104 loc) · 5.14 KB

cardhovernet_tutorial.md

File metadata and controls

131 lines (104 loc) · 5.14 KB

Running a SpatialOne Analysis using CARD & Hovernet

For this tutorial, we assume the user is using the following path and has Docker properly set up: /home/user/card_demo. It is also assumed that the user has access to a machine with a GPU and the right memory set up.

1. Retrieve SpatialOne's Code

Donwload the code from the github repository

git clone https://github.com/Sanofi-Public/spatialone-pipeline.git
cd spatialone-pipeline

2. Download Required Data

2.1. Sample Data

Download the two cancer samples, the reference data for deconvolution, and the default configuration files.

./download_experiment_data.sh

2.2. Downlaod Hovernet Weights

Hovernet can work with different weights, as described in their official repository. To use the CONSEP weights and labels described in the SpatialOne paper you should download them on the reference folder.

You can manually download them from the author's site, from Zenodo (weights, labels) or you can execute the following two instructions:

curl -L -o reference/hovernet_original_consep_type_tf2pytorch.tar "https://zenodo.org/records/12801948/files/hovernet_original_consep_type_tf2pytorch.tar?download=1"
curl -L -o reference/type_info.json "https://zenodo.org/records/12801948/files/type_info.json?download=1"

3. Set Up the Environment

Prepare your environment to build the container images if necessary. To that end, edit the .env file:

HOST_DATA_PATH = /home/user/card_demo/spatialone-pipeline
GPU_DEVICE_ID = 0

Export proxies only if necessary:

export HTTP_PROXY=http://your.proxy.url:port
export HTTPS_PROXY=http://your.proxy.url:port

4. Build the Docker Image

Build the docker containers:

make docker-build

Note that this operation may take up to 20 minutes

5. Edit the Configuration Files

Edit the config file located at $HOST_DATA_PATH/conf/visium_config_flow.yaml (in our case ./conf/visium_config_flow.yaml) to reflect an appropriate configuration. You'll need to define the pipelines to run, and set up Hovernet & CARD.

5.1 Enable All Pipelines

Ensure that all the pipelines are enabled:

pipelines.enabled:
    imgseg: True     # cell segmentation
    cell2spot: True  # matching cells to visium spots
    celldeconv: True # cell deconvoluiton
    cluster: True    # morphological clustering
    assign: True     # cell assignment integrates celldeconvolution with cell segmentation
    qc: True         # QC metrics generation
    datamerge: True  # To visualize in Tissuumaps enable "datamerge: true"
    spatialanalysis: True # Spatial analysis reporting

5.2 Enable Hovernet Configuration

Remove the cellpose configuration section and uncomment the hovernet section:

imgseg:
    image.resolution: "400dpi"
    image.magnification: "20x"
    model:
        name: "hovernet"
        version: "0.0.1"
        params:
            gpu: '0'
            nr_types: 5
            batch_size: 16
            model_mode: "original"
            nr_inference_workers: 1
            nr_post_proc_workers: 1
            wsi: True
            tile: False
            help: False
            model_name: "hovernet_original_consep_type_tf2pytorch.tar"
            type_filename: "type_info.json"

Please note that if you plan to use a different set of weights than hovernet_original_consep_type_tf2pytorch you should update the model_name and the type_filename parameters.

5.3 Set Up CARD Configuration

Remove the cell2location section and uncomment the CARD block:

celldeconv:
    model:
        name: "card"
        version: "X.X.X"
        params:
            min_count_gene: 100
            min_count_spot: 5
            ct_varname: 'cellType'
            ct_select: 'NULL'
            atlas_type: 'luca'
            sc_label_key: 'luca'
            sc_sample_key: 'batch'

Ensure to update the atlas_type parameter to luca to match the reference dataset with the type of tissue analyzed in the example. If you plan to use a different reference dataset, make sure to set up the appropriate refernce dataset name.

6. Run the Analysis

Run the analysis:

make docker-start

By following these steps, you will set up and run a SpatialOne analysis using CARD and Hovernet.

7. Results Visualization

The analysis results can be loaded in TissUUmaps by loading the experiment_tmap.tmap file in the TissUUmaps application (File --> Open --> select the appropriate file): image

Or by checking the html reports stored at results/experiment_name/reports image