AB labeling

Our project on understanding control of autophagosome size and number requires the segmentation of autophagic bodies (ABs) in hundreds of TEM images. To increase the speed and reproducibility of this process, we have generated a Cellpose 2 model that can automatically label APBs in our images. Thanks Emily Marron and her mentor Dr. Andrew Ross for doing most of the work training and validating this model, and to Jonathan Backues for labeling of most of the training images. Here's a link to the model: Cellpose Autophagic Body model (t_2_201)

This work has now been published in the journal Autophagy: Emily C. Marron, Jonathan Backues, Andrew M. Ross and Steven K. Backues (2024) Accurate automated segmentation of autophagic bodies in yeast vacuoles using cellpose 2.0 Autophagy https://doi.org/10.1080/15548627.2024.2353458 or Download PDF of author's accepted manuscript

Here's an example of what this model can do:

Here's some instructions for use:

1. Acquire TEM images of APB in yeast vacuoles, carefully following the steps described in section 2.1 of Backues et al. 2014

2. Manually sort out images of vacuoles that do not contain any bodies. These will still be used for the analysis of APB number, but do not need to be segmented.

3. Download the APB model (above)

4. Download, install and run Cellpose2 to segment the images containing APBs

a. For simplest use, install the Cellpose2 GUI using instructions from https://github.com/MouseLand/cellpose. The Cellpose2 GUI allows images to be segmented one at a time: Load the APB model, load the desired image, choose “run model” to generate masks, and then save these as png/tif.

b. For those with more computational experience, automated segmentation can be performed faster in batch using Pycharm and a Docker container (biocontainers/cellpose:2.1.1_cv2); utilities for this can be found at https://github.com/StevenBackues/cellpose_APB.

5. Copy the mask files to their own folder, and use these to measure the area of each labeled APB and the number of APBs per vacuole. This can be easily done in python using code provided above or in ImageJ using the MorphoLibJ library.

6. Estimate the original size and number distribution of the bodies from this data following the instructions in section 2.3 of Backues et al. 2014

Original data, including the images used for training and testing, can be found on OSF at https://osf.io/tuhwn/.

Utilities for batch processing with cellpose 2.0 and a python script for measuring the resulting masks can be found on Github at https://github.com/StevenBackues/cellpose_AB

Page updated

Report abuse