Skip to content

Mammary cell shape computation over time with Trackmate-Cellpose

License

Notifications You must be signed in to change notification settings

cadaws/MaSCOT-AI

Repository files navigation

MaSCOT-AI

Mammary cell shape computation over time with Trackmate-Cellpose

This code was used to analyse intravital microscopy data of single mammary gland cells in mice in puberty, adulthood and pregnancy.
Associated publication: Dawson, Milevskiy et al. Cell Reports, 2024, Hormone-responsive progenitors have a unique identity and exhibit high motility during mammary morphogenesis.

Deposited model, data, training images and example analysis

The MaSCOT-AI Cellpose model has been deposited on Zenodo (10.5281/zenodo.14503476) 2D single channel maximum projections from 4D intravital movies have also been deposited on Zenodo, along with the extracted 5th time points used for training and the segmentation files. I will also upload a folder of example movies and the resulting data exported from Trackmate-Cellpose.

Data acquisition

Movies were acquired using an Olympus-FVMPRS system with a 25x objective and 1.5 zoom at 512x512 pixels and 2 µm z-step. Custom filters for Confetti fluorescent protein detection are described in the paper.
See our protocol also:

Data processing

The original 4D movies were stabilised by a combination of 4D cell tracking and 'Correct 3D drift' in Imaris in FIJI with HyperStackReg 350 2D movies of single cell spans (10-15 µm) were generated manually by 3D crop in Imaris, then flattened by maximum projection in FIJI

Model training

The MasCOT-AI model was trained in Cellpose 2 on 150 still images (the fifth time point from a representative sample of movies). Training was initialised with the CP model and automatic cell size determination.

Trackmate-Cellpose analysis

Analyse 2D movies with the Cellpose model within Trackmate to connect cell measurements over time, then extract and visualise meaningful information.

Python script: Trackmate-Cellpose MaSCOT-AI.py
Incorporates the MaSCOT-AI model into Trackmate and loops through tiff time lapses to export various cell shape descriptors as csv
Based on:
https://imagej.net/plugins/trackmate/scripting/scripting
https://imagej.net/plugins/trackmate/scripting/trackmate-detectors-trackers-keys
https://github.com/trackmate-sc/TrackMate-Cellpose

R script: 1 Read and split Trackmate-Cellpose data MaSCOT-AI
Compiles all cell timelines for each measurement in large nested lists

R script: 2 Unnest and transform Trackmate-Cellpose data MaSCOT-AI
Rearranges the lists into a simple table so that each row represents a cell with an ID, sample label and measurements over time.

Excel file: MaSCOT-AI_ellipse_aspectratio_analysis
Example of data analysis including:

  • Extraction of cell tracks representative of percentiles
  • Running average calculation
  • Measuring peak frequency

Thanks for your interest and please get in touch @calebadawson (X/Twitter)