Calibrate Marginalization
This allows you to replicate the analysis from our most recent paper (pre-print). It allows you to extract lineage dynamics from uncorrected but filtered data in a statistically rigorous manner using survival analysis.
Step 1: Detect missed divisions
In survival analysis it is key that the chance of the event under study (in our case cell divisions) is independent from the chance of being lost to follow-up. During cell tracking cells are often lost just before division which violates this condition. To break the relationship between cell division and being lost, we need to look a the end of every track fragment and evaluate if it was in the process of dividing.
We do this with a neural network that detects if the cell is in the division process, defined as three frames before and after the moment of chromosome separation. This network can be trained the same as the normal division detection network by setting full_window to True in the organoidtracker.inifile.
The first step is thus to run the full window division network on the filtered data. A pre-trained network trained on organoid data is available here.
Step 2: Run analysis in GUI
You can now make the survival curves using use Tools -> Cell cycle -> Survival curves .... If you are in the all_experiments tab it will automatically make curves for all organoids.
Step 3: Export data
For more in depth analysis you can export the cell cycle lengths using File -> Export cell cycle info. If you want to customize the output you can adapt plugin_cell_cycle_exporter.py.