This repo contains a minimal example demonstrating how to compute a simple metric (in this case, mean marker expression) by cell type using the published results from the CODEX cytokit+sprm pipeline from the HubMAP data portal.
The analysis requires three inputs:
- The multiplexed image from the CODEX Cytokit+SPRM pipeline (from globus)
- The corresponding mask produced by the CODEX Cytokit+SPRM pipeline (also from globus)
- The cell type predictions
The celltype predictions are not currently publicly available as the model is still under development. However, a module has been provided which integrates with the Cytokit+SPRM pipeline to directly produce celltype predictions. This is availble to consortium members at https://github.com/hubmapconsortium/deepcelltypes-hubmap.
The CWL workflow referenced above adds one more output to the CODEX Cytokit+SPRM
pipeline: a .csv file called deepcelltypes_predictions.csv containing the cell type
predictions with the following structure:
mask_index,centroid_x,centroid_y,predicted_celltype
First, create a virtual envrionment and install the dependencies with
pip install -r requirements.txt
Next, make sure you have all of the necessary CODEX Cytokit+SPRM data downloaded from globus for the dataset you are working with.
Finally, make sure you have the celltype predictions. These are computed either by running the Cytokit+SPRM pipeline with the additional prediction CWL workflow referenced above, or by unpacking the predictions that I've shared with consortium members from the preliminary model development.
Modify the pipeline_output_location, celltype_prediction_location,
and dataset values in mean_expression_example.py to match the locations
on your system.
Then:
python mean_expression_example.py