• aims to create a simple, extensible, and reproducible pipeline to segment (or infer) and quantify the shape, position, size, and interaction of multiple intracellular organelles from confocal microscopy 🔬 images.
• is modular 🔢 to support a diversity of organelle-focused research questions.
• can be applied broadly to many types of in vitro 🧫 and in vivo models 🐁🧬 to better understand the spatial coordination and interactome of organelles during key biological processes or disease.

Compatable Organelles 🔓🗝️

• Lysosomes
• Mitochondria
• Golgi
• Peroxisomes
• Endoplasmic reticulum
• Lipid droplets
• Cell
• Nucleus

Outside segmentation methods can also be used to incorporate additional oragnelles.

infer-subc and the related organelle-segmenter-plugin for Napari are available on PyPI via:

pip install infer-subc
pip install organelle-segmenter-plugin

A full list of dependencies and recommended setup steps are included in env_create.sh.

NOTE: Proceed to the Organelle Quantification section below if you have already created instance segmentations using a separate method.

The starting point for the infer-subc analysis pipeline is to perform instance segmentation on single or multichannel confocal microscopy images, where each channel labels a different intracellular component. Each organelle will be segmented from a single intensity channel. The subsequent analysis is performed at a single-cell level, so we have also developed several workflows to segement the cell area.

We recommend that you use our infer-subc implementation for Napari called organelle-segmenter-plugin for segmentation. This will allow users to systematically test segmentation settings for each organelles, then batch process all organelles of interest across multiple cells. Alternatively, you can utilize the included set of Jupter Notebooks or your own python script calling functions from the infer_subc module to work through the segmentation process step by step for each cell.

You must have installed organelle-segmenter-plugin as described above to use this method. The order in which you run the workflows is not important, but for down stream quantification, you must include at least one organelle and the cell mask.

1. Open Napari. Then drag-and-drop or use the File > Open File(s)... controls to open a single- or multi-channel confocal microscopy image. This image will be used to test the segmentation settings you want to apply to many cells during batch processing.
2. Start the plugin by navigating to Plugin > Infer sub-Cellular Object Npe2 plugin > Workflow editor. A right side panel will appear.
3. In the Workflow editor, select the image to work on.
4. Select the workflow corresponding to your desired organelle(s)/masks.
5. Adjust the parameters for each step based on the intermediate results. If you need to return to a previous step, you must restart the workflow by pressing Close Workflow at the bottom of the panel.
6. Save the workflow settings that work for your image by using the Save Workflow option at the bottom of the panel. Save each of the settings files to process a single image into the same location for batch processing. IMPORTANT: the file name should end with the same name as the workflow you are working on.

   Naming Examples:

   For settings saved from the 0.2.lyso workflow, the following naming are acceptable:

   • "20241031_lyso.json"
   • "iPSCs_lyso.json"
   • "lyso.json"

   Do NOT use names like:

   • "lysosomes.json"
   • "LS.json"

7. Once all of the settings are saved, open the batch processor by going to Plugins > Infer sub-Cellular Object Npe2 plugin > Batch processing.
8. Load the saved workflow settings for all channels and specify the input (intensity images) and output (desired location for segmentation files) folders.
9. Click Run. A progress bar will allow you track your processing.

The primary purpose of the Juptyer notebooks are to walk step-by-step through each of the segementation workflows. We hope these notebooks provide a more easily accessible resource for those who are new to Python image analysis. The notesbooks below include sets to segment a single image. A batch processing workflow has not yet been created.

Step 1️⃣: Identify a single cell of interest

Segment (or infer) the cellmask and nuclei from an image depending on the type of fluorescent labels used and the number of cells per field of view (FOV):

• Fluorescently labeled nuclei (no cell or plasma membrane)
  • one or more cells per FOV
• No cell, plasma membrane, or nuclei labels with
  • one cell per field of view FOV
  • one or more cells per FOV

Step 2️⃣: Segment organelles

Each of the organelles you wish to include in your analysis should be segmented from a single fluorescently labeled structure.

2. Infer lysosomes
3. Infer mitochondria
4. Infer golgi
5. Infer peroxisomes
6. Infer endoplasmic reticulum (ER)
7. Infer lipid droplets

After processing all cells in your dataset, we recommend you quality check your segmentation results by visually inspecting the images. The Segmentation Validation pipeline is included in the Full Quantification Pipeline Notebook to streamline the validation process.

🚧 In a future verions, this notebook will also include quality checks for assumptions made during quantification (i.e., only one nucleus and ER per cell, etc.).

After each of the organelles of interest are segmented, single or multi-organelle analysis can be carried out using Jupyter Notebook-based pipeline(s). Each of the following analysis types are modular and can be used in combination or separately.

Combined Analysis:

• Full Quantification Pipeline: quantification of the morphology, interactions, and distribution of any number of organelles within the same cell. This pipeline incorporates batch processing within a single experiment and summarizes the results across multiple experimental replicates.

Individual analysis pipelines :

The following notebooks primarily act as a step-by-step guide to understanding each measurement type. However, they can also be used to quantify features of single organelles or pairs of organelles (interactions) from individual cells.

• Organelle morphology
• Pairwise organelle interactions
• Subcellular distribution
• Cell/nucleus morphology
• COMBINED ANAYSIS ONLY: batch processing
• COMBINED ANAYSIS ONLY: per-cell summary

🚧 Future implimentations of these notebooks will include batch processing capabilities (e.g., multiple cells, multiple organelles) for each quantification type separately.

A quick note on tools and resources used...

• napari-allencell-segmenter -- We are leveraging the framework of the napari-allencell-segmenter plugin, which enables powerful 3D image segmentation while taking advantage of the napari graphical user interface.
• aicssegmentation -- We call the aicssegmentation package directly to access their advanced segmentation functions.
• napari -- Used as the visualization framework, a fast, interactive, multi-domensional image viewer for Python.
• scipy -- Image analysis
• scikit-image -- Image analysis
• itk -- Image analysis
• numpy -- Under the hood computation
• 'pandas' -- Quantitative data manipulation

Early in the develepmont we chose to leverage methods created in the Allen Cell & Structure Segmenter and napari plugin. Although the logic of our multi-channel organelle segmentations required us to fork and modify their code, we hope it porvides a stable, but evolving base which will help manage accumulation of technical debt. In addition to the overall logic, we particulary leverage their workflow paradigm which is integral in the use of the napari plugin interface. Implementation of infer-subc as a Napari plugin using this framework is called organelle-segmenter-plugin.

​The Allen Cell & Structure Segmenter is a Python-based open source toolkit developed at the Allen Institute for Cell Science for 3D segmentation of intracellular structures in fluorescence microscope images: aicssegmentation package. This toolkit brings together classic image segmentation and iterative deep learning workflows first to generate initial high-quality 3D intracellular structure segmentations and then to easily curate these results to generate the ground truths for building robust and accurate deep learning models. The toolkit takes advantage of the high replicate 3D live cell image data collected at the Allen Institute for Cell Science of over 30 endogenous fluorescently tagged human induced pluripotent stem cell (hiPSC) lines. Each cell line represents a different intracellular structure with one or more distinct localization patterns within undifferentiated hiPS cells and hiPSC-derived cardiomyocytes. Here, we leveraged select segmentation methods specialized for a particular organelle shape (i.e., round versus tubular objects) to carried out segmentation states within the infer-subc segmentation workflows.

If you encounter any problems, please file an issue with a detailed description.

Read the CONTRIBUTING.md file.

Distributed under the terms of the [BSD-3] license.

infer-subc and organelle-segmenter-plugin are free and open source software.

• CZI Neurodegeneration Challenge Network (NDCN)

infer-subc analysis has been featured in:

1. Shannon N. Rhoads, Weizhen Dong, Chih-Hsuan Hsu, Ngudiankama R. Mfulama, Joey V. Ragusa, Michael Ye, Andy Henrie, Maria Clara Zanellati, Graham H. Diering, Todd J. Cohen, Sarah Cohen. Neurons and astrocytes have distinct organelle signatures and responses to stress. bioRxiv 2024.10.30.621066; doi: https://doi.org/10.1101/2024.10.30.621066