Visualize tracked cells and their features

[ziw-liu]

Visualize TCZYX images and TYX tracking labels.

Loads T(N*Z) UMAP embeddings into the labels layer's attribute to be used with napari-clusters-plotter.

Hover on input fields to view tips.

Before testing

Install again with pip install -e ".[dev]".

Apple silicon users must install higra from conda-forge before installing this plugin. Upstream issue: higra/Higra#249

[mattersoflight]

@ziw-liu can you point me to paths of matching zarr stores so I can try this out?. Found them in code.

[mattersoflight]

@ziw-liu installing napari-clusters-plotter via mamba as per the README of project breaks the napari environment. pip install napari-clusters-plotter plays well with our other instructions on Wiki.

[mattersoflight]

The loading crashed while trying the FOV B/4/0 and is quite slow with B/4/2 on the login-02 node. Slowest parts are loading the movie and, somewhat surprisingly, loading features.

What contributes to the slow loading of features? Can we lazy load time points, given the typical XYZ chunks?

[mattersoflight]

@ziw-liu I now see a peculiar behavior. The INFO logs say that the data is being loaded, but the canvas stays empty. Let's take a look at this in person.

[ziw-liu]

@mattersoflight a know issue is that napari-clusters-plotter does not work with the new napari 0.5.0: BiAPoL/napari-clusters-plotter#308 (comment). However 0.5.0 brings useful features for users of other widgets so I didn't pin the version. I'm also seeing flakiness on HPC with the also newly released PyQt 6.7.1, sometimes using PyQt5 seems better.

[ziw-liu]

@mattersoflight Now if you do:

pip install -e ".[dev]"

This will include [all] that installs napari, pyqt5, and napari-clusters-plotter.

[mattersoflight]

@ziw-liu works well. I enjoyed watching cells move from one part of UMAP space to another. The viewer widget can use a few tweaks (after the current work on data formats and VisCy is completed):

• show tracks (via napari.view_tracks().
• rename the fields for clarity.

[Soorya19Pradeep]

@ziw-liu, I was able to use the plugin as well. I can see the whole track of the cell starting from timepoint 0. Is it possible to visualize the progression of tracking along time instead?

I think I can see the progression sliding over the z-slice slider though.

[ziw-liu]

@Soorya19Pradeep trajectory dimensionality has been fixed.

[mattersoflight]

@ziw-liu the updated field names read well.

Following improvements are needed:

• dependencies for the plugin need to be worked out and documented. I needed to install the following:

  • iohub from github (main branch)
  • ultrack
    Since this plugin is experimental, why not install iohub from github?
    Although napari-clusters-plotter requires numpy ~<1.26, it worked just fine with numpy 2.0.2 installed by ultrack.

• Improve the analysis workflow:

  • compute the 3-component UMAP and 6-component PCA across all FOVs in advance to enable comparison of cells across FOVs.
  • rename the widget to view embedding dynamics.
  • our widget loads the napari-clusters-plotter/plotter-widget with correct layer selected. We don't need napari-clusters-plotter/dimensionality-reduction.

• Clear the layers when Load button is pressed. If layers are present, and the Load button is pressed, I get an error.

  Signal emitted at: /hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/magicgui/widgets/bases/_value_widget.py:71, in _on_value_change
    >  self.changed.emit(value)

  Callback error at: /hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/napari/layers/_scalar_field/scalar_field.py:381, in _get_level_shapes
    >  shapes = [self.data.shape]

    Local variables:
       data = [array([[[[547.1515 , 524.1734 , 539.83496, ..., 572.00104, ...

[mattersoflight]

Animations like the following are handy for exploring each FOV. The speed of analysis will improve if we don't have to do the dimensionality reduction for each FOV.
https://github.com/user-attachments/assets/97fc8824-622b-44ff-b324-212932fc08b3

[mattersoflight]

I couldn't cluster the instances manually - perhaps because I have napari 0.5. I get the following error when I draw an ROI.

File "/hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/napari_clusters_plotter/_Qt_code.py", line 516, in onselect
    self.parent.manual_clustering_method(self.ind_mask)
  File "/hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/napari_clusters_plotter/_plotter.py", line 125, in manual_clustering_method
    self.run(
  File "/hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/napari_clusters_plotter/_plotter.py", line 730, in run
    self._update_cluster_image(
  File "/hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/napari_clusters_plotter/_plotter.py", line 805, in _update_cluster_image
    self.visualized_layer = self._draw_cluster_image(
                            ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/napari_clusters_plotter/_plotter.py", line 862, in _draw_cluster_image
    cluster_layer = Layer.create(
                    ^^^^^^^^^^^^^
  File "/hpc/mydata/shalin.mehta/envs/ndutils/lib/python3.11/site-packages/napari/layers/base/base.py", line 2310, in create
    raise TypeError(
TypeError: _add_layer_from_data received an unexpected keyword argument ('color') for layer type labels

[ziw-liu]

dependencies for the plugin need to be worked out and documented.
perhaps because I have napari 0.5. I get the following error when I draw an ROI.

@mattersoflight I fixed it in 6d1b8f4. Did you run installation again after pulling?

[Soorya19Pradeep]

@ziw-liu , I am getting the following error after I pulled the changes:

File /hpc/mydata/soorya.pradeep/tracking/lib/python3.10/site-packages/npe2/_command_registry.py:34, in CommandHandler.resolve(self=CommandHandler(id='napari-iohub.get_reader', fun...on_name='napari_iohub._reader:napari_get_reader'))
32 self.function = utils.import_python_name(self.python_name)
33 except Exception as e:
---> 34 raise RuntimeError(
self.python_name = 'napari_iohub._reader:napari_get_reader'
self = CommandHandler(id='napari-iohub.get_reader', function=None, python_name='napari_iohub._reader:napari_get_reader')
35 f"Failed to import command at {self.python_name!r}: {e}"
36 ) from e
38 return self.function

RuntimeError: Failed to import command at 'napari_iohub._reader:napari_get_reader': No module named 'iohub.ngff.nodes'; 'iohub.ngff' is not a package

Any idea what happened here?

[ziw-liu]

@Soorya19Pradeep Please run pip install again after pulling. There were some dependency updates to make ultrack work.

[Soorya19Pradeep]

Thanks @ziw-liu ! It works well!

[Soorya19Pradeep]

I have tested visualizing tracks and features for the June dataset using these changes. Works really well!