Celldetective

- Introducing checkable well & position lists to better control the conditions to process or redo with new parameters as well as more intuitive selection of data for table representations
- Automatic check on startup for a newer version of the software (a message in the terminal tells you to update if you are not up-to-date)
- Small hover response on QCheckBox options
- New measurements are available for background normalized images (typically RICM/IRM/BF); you can now measure the number of pixels < 1 in the mask and estimate the fraction of the mask area that it represents
- Auto correct missing labels upon measuring
- Bug corrected in the event detection annotator with the update of class/status colors after saving
- Feat: the selected cell remains selected when switching from one event class to another in the annotator
- Prevent negative values from entering square root operator in filters
- Texture computation is now skipped on completely black frames to save time

**Full Changelog**: https://github.com/celldetective/celldetective/compare/v1.3.3.post1...v1.3.4.post1

- Update the "How to cite" with the latest version of the Celldetective manuscript (https://www.biorxiv.org/content/10.1101/2024.03.15.585250v3)
- Complete redesign of the spot detection viewer, to inherit from the StackVisualizer class. Real-size scatter plot showing detections instead of adding Circle patches.
- Spot detection: switch to LoG method, more robust. Refactoring of the code to make computation faster. Store spot counts and average intensity of spots. Replace with newest spot detection upon remeasuring.
- Replace with newest texture measurements upon remeasuring.
- Automatically delete corrupted segmentation models that do not have a config_input.json file (they can't be opened by Celldetective).
- Prompt Cellpose/StarDist parameters every time a generalist model is called, to be able to switch from a model to another.
- Automatically add a newly trained model to the Model zoo and refresh to display it.
- Single images can now be processed, with intelligent detection of the number of channels and axes (to separate A 3-channel image from 3 frames with 1 channel)
- Automatic padding of segmentation annotation crops that are too small for StarDist (less than 256*256 pixels)

**Full Changelog**: https://github.com/celldetective/celldetective/compare/v1.3.2...v1.3.3.post1

* Table exploration for cell pairs: new collapse-in-neighborhood function to collapse the measurements per cell pair or reference cell, including or excluding non-contact states
* Function `expand_pair_table` to merge pair tables and single-population tables (in relative_measurements.py), also replaced in GUI
* Function to extract a specific neighborhood data in a pair table `extract_neighborhood_in_pair_table` (in neighborhood.py)
* auto-scaling in classifier widget that adds a 5% pad on measurements values to not cut exactly at min/max values in -x and -y
* Threshold event detection: new safety net to prevent the detection of times larger than the stack length due to an approximate fit of the binary response function (truncated automatically to the max frame) (from DizBell request)
* Export survival data: per well or position according to the plotting set up, either full Kaplan-Meier estimate with confidence interval or a single-timepoint survival or any ECN% (as in EC50) as a table with the experiment metadata (concentrations, antibodies, cell types etc) ready for plots or export (request by DizBell)
* Table UI: new line plot function from seaborn, works particularly well when passing `FRAME` as -x to view the average response (non-synchronized) of a population, or the average change over time of a quantity (e.g. fluo)
* Table UI: features to take the LOG10 of a quantity (values <= 0 yield NaN), and calibrate a measurement with a multiplicative factor
* Fix: replace with the newest measurements when remeasuring a quantity in Celldetective (before, the oldest was kept)
* Fix: bug with the mask-contact neighborhood method, yielding wrong reference-neighbor pairs
* Feat: new mask intersection quantity for mask-contact method to evaluate how much target/effector masks are superimposed. Expressed relative to target or effector area when area has been computed before interaction study. Intersection of 0 means lateral contact but no actual superposition (eg: contact within a population, sharing same label_targets/effectors, will always give 0 for touching cells)
* Time series collapse module: include class 2 cells in "all cells" plotting option.

Contributors
==========

* remyeltorro

**Full Changelog**: https://github.com/celldetective/celldetective/compare/v1.3.1...v1.3.2

* New preprocessing feature to add lateral offsets to specific channels, following request 11 : the feature behaves like background corrections, the user defines a protocol list (that can mix background corrections and channel offsets)
* Skip preprocessing corrections on black (empty) frames for performance
* Measure intensity of NaN for completely black frame (void of information): channels with only endpoint information now measure NaN everywhere except at the endpoint
* No scatter rescaling when applying classification in classifier widget to address a request from DizBell
* New option to exclude late events from survival study, request from DizBell and laurentlimozin
* ADCC demo was mistakenly removed from Zenodo; it has been reintroduced
* New lysis detection model based on PI intensity + nuclear area (lysis_PI_area event detection model)
* Bug fixes

Contributors
==========

* remyeltorro

**Full Changelog**: https://github.com/celldetective/celldetective/compare/1.3.0.post1...v1.3.1

- Bug fix in the classifier widget: the class name was not updated properly, preventing color classification and writing the wrong information in the tables
- Load incomplete labels when viewing cell tracks in napari

**Full Changelog**: https://github.com/remyeltorro/celldetective/compare/v1.3.0...1.3.0.post1

Introducing statistical tests and effect sizes

It is now possible, for the first time in Celldetective to perform the Kolmogorov-Smirnov statistical test between all combinations of x+hue distributions tested when plotting boxplots or histograms in the table exploration module. Similarly, we embedded the computation of Cliff's Delta, a non parametric effect size estimate. The p-values / effect size are represented as a color coded pivot table with the first distribution as row.

New datasets and models

Spreading assays are now associated to a dataset of spreading immune cells (primary natural killer, mice T cells) and an accompanying Cellpose model, based exclusively on RICM.

The MCF7 models have been udapted and the dataset cleaned up.

More model/dataset updates will be rolling in the next few weeks.

Notable changes

- In the classifier widget, cells are interpreted as False concerning the condition before their arrival (and not NaN like before). As an example, a spreading cell classified as spread with a condition on the RICM intensity (intensity < 1) is evaluated to False before it appears on the image, until it crosses the threshold
- Classification functions removed from GUI and migrated to measure.py to be usable in scripts
- It is now possible to open a partially segmented stack/segmentation in the napari-segmentation viewer. The missing label frames are passed as black/empty frames
- Ignore bTrack dummies when relabeling segmentation masks (that were taking existing values)
- Interpolation of image NaNs before texture computation to prevent crash
- Better panel (PROCESS, INTERACTIONS) expansion response (no window resizing if not needed)
- Transparency slider for single-cell signal representation
- Restructuring of the plot survival / plot signal modules to use the same UI blocks
- Survival computation removed from GUI and migrated to events.py to be usable in scripts
- Bug fixes

Contributors

- remyeltorro

**Full Changelog**: https://github.com/remyeltorro/celldetective/compare/v1.2.2.post2...v1.3.0