This Python package, MitoScripts, assists in quantifies mitochondrial morphology. It was written by me, Grant Hussey, as a rotation project in the Holt Lab at NYU Langone Health's Institute of Molecular Biomedicine.
This package first requires data to be processed into 3D models by using MitoGraph, an open-source software developed by Matheus Viana.
This build of MitoScripts works for MitoGraph v3, and was tested on mammilian cells.
MitoGraph is a software that takes z-stack images of mitochondria and produces a 3D model. This Python package, MitoScripts, takes the output from MitoGraph and quantifies it.
Be sure to have a consistent naming scheme for these image files that do not repeat, in order to automate the quantification process downstream.
2. Process your data through MitoGraph.
3. Run MitoScripts.
After MitoGraph processing is completed, you get .gnet and .mitograph files. These are used by MitoScripts to quantify mitochondrial morphology.
First, gather the .gnet and .mitograph files into a single directory. Next, import the python package, which has three modules, mitographer, mitodata, and mitopca. You will most likely only interface with mitographer.
As of now, MitoScripts works best if used in scientific mode in an IDE or a jupyter notebook.
from mitoscripts import mitodata as mt
from mitoscripts import mitographer as mtgrapher
from mitoscripts import mitopca as mtpca
4. Run the analysis of choice.
This is the most self-explainatory MitoScripts module. Interface with it to produce automated graphs based on your MitoGraph outputs, .gnet and .mitograph files.
create_graph_suite takes data_dir, data_name, name_dict, and savefigs. Produces a scattered box plot graph for a plethora of quantifiable metrics from MitoGraph. This function returns a Pandas DataFrame containing the extracted metrics. Keep this DataFrame for downstream analysis.
data_dir: the directory of your .gnet and .mitograph files.data_name: a user-defined name unique to the experiment (for example, "20oct19_treatment_with_chemical_x")name_dict: a user-defined dictionary that maps your filenames to their respective labels.savefigs: set this toTrueto save figures inside of your current directory.
pca_suite takes df, or the DataFrame returned by create_graph_suite, and runs a PCA analysis. It also creates a scree plot and a centroid plot (where the centroid of each cluster is graphed instead of the points). It takes df, to_drop, and style.
-
df: as explained above -
to_drop: defaults to None. If None, will initialize to the automated output ofget_default_col_to_drop. This removes 1) redundant metrics and 2) irrevelant metrics. Set to a user-defined list if you want to customize the metrics used by MitoPCA. -
style: defaults to "tableau-colorblind10", a Seaborn style, as defined in the global variableDEFAULT_STYLE.
scatter_length_distribution takes data_dir, data_name, name_dict, and savefigs. Variable definitions are the same as above. This function will create a heatmap that displays the individual distributions of each image to visualize distribution of lengths in each image's greater mitochondrial network.
This is the main backbone of the MitoGrapher module. Of note, there may be one function you'd interface with:
analyze_images takes data_dir, name_dict, and data_name and produces the df explained above.
This module is the backbone of mitographer.pca_suite. There is one notable function:
run_special_pca uses an algorithm to drop columns producing the most dissimilarity in the data, then displays the resultant data's PCA graph. It takes df, threshold, min_cols.
threshold and min_cols work in tandem. They define two different conditions that the algorithm will stop. threshold defines the max cutoff for PC 1's explained variance ratio, while min_cols defines the min number of columns in df you wish to keep in the final DataFrame.
For example, let's say I have a DataFrame with 20 columns. The algorithm will start deleting columns that produce the most dissimiliarity. If deleting a column makes PC1's explanied variance ratio greater than threshold, it will stop deleting columns, then run a final PCA.
If threshold is not met, then the loop will continue.
Simiarly, the algorithm checks each loop if you've whittled the DataFrame down below min_cols. If so, the algorithm will also stop deleting columns and run a final PCA.
name_dict has key:value pairs as filename:treatment_name. For example, lets say you're treating cells with hydrogen peroxide and seeing if there is a change in mitochondrial morphology. You need to save your images in the form of "control_000", "control_001", "control_nnn" and "h2o2_000", "h2o2_001", "h2o2_nnn", then create a name_dict like so:
name_dict = { 'control':'Control Group', 'h2o2':'With Hydrogen Peroxide' }
This way, your graphs and variables will be displayed with the nicer headers "Control Group" and "With Hydrogen Peroxide."
For Step 1: Read the papers on MitoGraph (namely this one) to follow best imaging practices for good MitoGraph results.
For Step 2: If you wish to use the normal MitoGraph pipeline, please see MitoGraph's github repo for procedure. You may skip to the next step.
Otherwise, if curious, I have provided here my own ImageJ macros based upon those supplied by Matheus that were useful in my pipeline. In order of use:
Tiff_and_MaxProfs.ijm: This macro takes a directory of .nd2 files, the output of our confocal microscope, and produces a .tif image for each file. Next, it take sthe .tif files and produces a new z-stack containing max projections of each image. Use case: In the MitoGraph pipeline, you need to create such a z-stack to assist in cropping out individual cells from each image.
CropCells_Complete.ijm: This macro takes a directory containing the .tif files from above, the max projection from above, and an ROISet.zip file containing ROIs for each cell (user-defined) and crops out the ROIs, creating one new .tif image per ROI. Use case: In my pipeline, I need not only a directory containing each one file per cell ROI, as is used in the normal MitoGraph pipeline, but also a directory of directories where each directory contains a single ROI. This extra "directory of directories" is used on the compute cluster offered at my institution.
createMontage.ijm: This macro creates a montage directly comparing the .png file output from MitoGraph (which displays the mitochondrial segmentation) to the original .tif for manual quality control. Use case: Post-MitoGraph processing, use this to see if any catastrophic segmentation errors occured.
Dependencies: 'numpy', 'pandas', 'scikit-learn', 'python-igraph', 'matplotlib', 'seaborn'. Please import these into your environment.