Purpose of this package is to apply mesmer segmentation to SMI CellComposite images and use those segmentations to recompute downstream outputs like expression matrices.
This package hasnt been uploaded to pip current (as of 5/5/22)
To use this function on commandline you want to clone this repository followed by
git clone https://github.com/Jacobog02/mesmer_segment.git # get repository
cd mesmer_segment # go there
pip install . # install local package to your python
If you are interested in modifying this package and rerunning pipelines live you must start a new python virtualenv, then call the setup script with the develop parameter. You can then pip install the package and when you run the command it recompiles on the fly
python3 -m venv /path/to/mesmer_env ## make the env
source /path/to/mesmer_env/bin/activate ## Spark up the env
cd /path/to/mesmer ## Go to the github repo locally
python3 setup.py develop ## Config package to recompile on the fly
pip install . ## Install the package
mesmer_segment --help ## Yields help screen from onfly compliation
In order to segment and visualize cell overlapys use the following command:
mesmer_segment -i /path/to/outs/ -v
In order to both segment and generate an expression matrix, polygon vertices, etc. use the following command:
mesmer_segment -i /path/to/outs/ --save-npz --build-outs
The outputs can then be loaded into Seurat (feat/imaging branch) with:
data <- ReadNanostring(
mtx.file = "/path/to/mesmer_exprMat_file.csv",
metadata.file = "/path/to/mesmer_metadata_file.csv",
molecules.file = "/path/to/mesmer_tx_file.csv",
segmentations.file = "/path/to/mesmer_polygons_file.csv",
type = c("centroids", "segmentation")
)
segs <- CreateSegmentation(data$segmentations)
cents <- CreateCentroids(data$centroids)
segmentations.data <- list(
"centroids" = cents,
"segmentation" = segs
)
coords <- CreateFOV(
coords = segmentations.data,
type = c("centroids", "segmentation"),
molecules = data$pixels,
assay = 'RNA'
)
obj <- CreateSeuratObject(counts = data$matrix, assay = 'RNA')
# select cells in segmentations and in expression matrix
cells <- intersect(Cells(obj), Cells(x = coords, boundary = "centroids"))
coords <- subset(x = coords, cells = cells)
# Add coords as an Image with name `SMI`
obj[['SMI']] <- coords
7/20/21:
- Updating segmentation input to process TIFF files from the
RawMorphologyImages/directory. - Mesmer function now accepts multiple parameters to the model for segmentation and overlap .jpg images to evaluate performance.
- Mesmer run in compartment = "both" mode as standard, downstream functions would explode with variable input currently.
- DEPRECIATED: '--save-npz' parameter, the function will always write mesmer output in compressed format as a standard.