cmIF fork: mplexable

• Author: engje, bue
• Date: 2020-11-01
• License: GPLv3
• Language: Python3

Description: mplexable is a fork from Jennifer Eng's original cmIF mplex_image software library (https://gitlab.com/engje/cmif). cmIF is a Python3 library for automated image processing and analysis of multiplex immunofluorescence images.

Source: the latest version of this user manual can be found at https://gitlab.com/bue/mplexable/-/blob/master/README.md

HowTo

HowTo - Installation Mplexable

Python version:

A cornerstone of mplexable is cellpose, which is used for segmentation. Cellpose requires at the moment of this writing python version 3.8. We set the python requirement for mplexable accordingly in the setup.py. You can check if these requirements are still true (https://github.com/MouseLand/cellpose/blob/master/environment.yml). If this has changes, please drop us a line, that we can adjust the setup.py file. Thank you!

Python environment:

We recommend installing mplexable in an own python environment. Iff you run miniconda (or anaconda) you can generate a mplexable python environment like this:

conda create -n mplexable python=3.8

You can activate the generated mplexable environment like this:

conda activate mplexable

CPU based installation:

1. Install some basics.

pip install ipython jupyterlab

2. Install torch.

Check out the pytorch side (https://pytorch.org/get-started/locally/), if you want to install torch with pip or from source, rather than with conda. Installing pytorch is enough. There is no need to install torchaudio and torchvision.

conda install pytorch

3. Install cellpose.

pip install cellpose

4. Install mplexable.

pip install mplexable
pip install aicsimageio[czi]  # if you deal with carl zeiss images.
pip install imagecodecs  # if you deal with miltenyi macsima images.

5. Initialize a local config file.

Open a python shell and type:

from mplexable import configure
configure.link_local_config()
exit()

The local mplexable configuration file can be found at: ~/.mplexable/config.py.

Use your favorite text editor to edit the mplexable config file.

Nvidia GPU based installation:

1. Install some basics.

conda install ipython jupyterlab

2. Install torch.

Note: For running touch on a GPU, you have to know which cuda toolkit version you have installed on your machine. How depends on your operating system. We leave it your homework to figure out how to do that.

Check out the pytourch side (https://pytorch.org/get-started/locally/), to figure out how to install the latest version of torch, for your os, python package distro, and cuda toolkit. Installing pytorch is enough. There is no need to install torchaudio and torchvision.

The final installation command will look something like below.

conda install pytorch cudatoolkit=nn.n -c pytorch

3. Install cellpose.

It is important that the GPU based version of pytorch is already installed, so that the CPU based version not get installed!

pip install cellpose

4. Install mplexable.

pip install mplexable
pip install aicsimageio[czi]  # if you deal with carl zeiss images.
pip install imagecodecs  # if you deal with miltenyi macsima images.

5. Initialize a local config file.

Open a python shell and type:

from mplexable import configure
configure.link_local_config()
exit()

The local mplexable configuration file can be found at: ~/.mplexable/config.py.

Use your favorite text editor to edit the mplexable config file.

HowTo - Update Mplexable

After an update, it is important to re-link your local mplexable configuration file.

1. Update mplexable.

pip install -U mplexable

2. Link the local config file.

Open a python shell and type:

from mplexable import configure
configure.link_local_config()
exit()

HowTo - Run a Data Extraction Pipeline

The data extraction pipeline jupyter notebooks are in the jupyter folder.

• mpleximage_data_extraction_pipeline.ipynb: most elaborate data extraction notebook.
• codex_data_extraction_pipeline.ipynb: notebook to process codex data.
• mics_data_extraction_pipeline.ipynb: notebook to process miltenyi macsima data.

Download the raw version of this notebook. Open then notebook in Jupyter to process your data.

For each step in the pipleine there is a detailed description in the notebook.

Besides, all functions have a docstring. To receive additional help for a particular function, in the python shell type:

help(function)

HowTo - Protein Marker

All biomarker used in the assay have to be specified in the local mplexable configuration file that can be found at: ~/.mplexable/config.py.

DAPI, quenching round, blank, and empty markers have to be specified in the es_markerdapiblank_standard variable.

Proteins have to be specified in the es_markerpartition_standard variable. With the protein name, you have as well to specify where the protein in the cell is expressed. Possible partitions are: nucleus, nucleus membrane, ring (which is the cytoplasm), and cell membrane. If proteins are expressed in more than one cell partition, then they can be specified more than once. This is biological knowledge that becomes important in the feature extraction step of the pipeline. To check where the proteins are exactly expressed, we recommend consulting the human protein atlas.

Proteins that are used for cell segmentation have to be specified as keys in the des_cytoplasmmarker_standard variable. These proteins are usually specific for a sudden cell type, or some cell types, but not all cell types. E.g. Ecad is expressed in cytoplasm of cancer cells and tonsil and can be used as cell segmentation marker. For one cell type, more than one segmentation marker can be defined. In the values part of the dictionary, all protein markers have to be specified that might be expressed in the cytoplasm of this particular cell type. E.g. CK5 might be expressed in the cytoplasm of cancer cells. You can define as many cell types as you want. This is biological knowledge that becomes important in the feature extraction step of the pipeline. All the protein markers mentioned in the des_cytoplasmmarker_standard have also to be specified as ring partition protein in the es_markerpartition_standard variable.

HowTo - Naming Convention

The naming conventions are specified in the local mplexable configuration file that can be found at: ~/.mplexable/config.py. The standard naming convention is influenced by our use of axioscan and the zen software (both from zeiss. It is totally possible to adjust this setting to own needs. The most crucial variables to look at are:

Basic setting:

• d_nconv['s_round_axio']: how is in the filename the staining cycle round defined.
• d_nconv['s_quenching_axio']: how is in the filename a quenching round specified.
• d_nconv['s_color_dapi_axio']: how is in the filename the actual microscopy channel specified.
• d_nconv['ls_color_order_axio']: a by wavelength sorter list of microscopy channel labels.
• d_nconv['s_sep_marker_axio']: character used in the filename to separate markes.

Czi image file name:

• d_nconv['s_czidir']: directory name where the raw czi images (with microscopy metadata) are stored.
• d_nconv['s_format_czidir_original']: string to specify the subfolder structure, e.g. one folder per slide, in the czi directory.
• d_nconv['s_regex_czi_original']: regular expression string to extract information from the czi filename.
• d_nconv['di_regex_czi_original']: dictionary to map the extracted information.

Raw tiff image file name:

• d_nconv['s_rawdir']: directory where the raw tiff files, one tiff file per round and marker, are stored.
• d_nconv['s_regex_tiff_raw']: regular expression string to extract information from the raw tiff filename.
• d_nconv['s_format_czidir_original']: string to specify the subfolder structure, usually one folder per slide, in the raw directory.
• d_nconv['di_regex_tiff_raw']: dictionary to map the extracted information.

HowTo - Slurm

All calculation intensive function offer the option to be processed as slurm job. This can be achieved by setting the parameter: s_type_processing = 'slurm'. The default setting is 'non-slurm'. Use the jupyter lab / Edit / Find ... and Replace All function, to replace them all.

However, the core slurm function, slurmbatch, is specified in the local config file that can be found at: ~/.mplexable/config.py. If you want run mplexable on a slurm cluster, you might have to modify this function, so that the resulting slurm command fits the scheme required by your super computer. The crucial parameter to look at and manipulated are: s_jobname, s_partition, s_gpu, s_mem, s_time, s_account.

Tutorial

The tutorial is here: https://gitlab.com/bue/mplexable/-/blob/master/TUTORIAL.md

Discussion

Mplexable is a very lightweight and flexible pipeline to process cyclic immunofluorescence images. Except for the registration code (which is Matlab script), mplexable is entirely written in Python3. To run mplexable you have to be savvy in the Python language, else you will struggle! To adjust the file naming convention standard in the config file and the registration function call, you have to be familiar with regex.

References

"A framework for multiplex imaging optimization and reproducible analysis." Eng J, Bucher E, Hu Z, Zheng T, Gibbs S, Chin K, Gray JW. Biorxiv: https://doi.org/10.1101/2021.11.29.470281

"Cyclic Multiplexed-Immunofluorescence (cmIF), a Highly Multiplexed Method for Single-Cell Analysis." Eng J, Thibault G, Luoh SW, Gray JW, Chang YH, Chin K. Methods Mol Biol. 2020;2055:521-562. https://doi.org/10.1007/978-1-4939-9773-2_24

"cmIF: A Python Library for Scalable Multiplex Imaging Pipelines." Eng J, Bucher E, Gray E, Campbell LG, Thibault G, Heiser L, Gibbs S, Gray JW, Chin K, Chang YH. In: Mathematical and Computational Oncology. ISMCO 2019. Lecture Notes in Computer Science, vol 11826. Springer, Cham. https://doi.org/10.1007/978-3-030-35210-3_3