DataLoader about ncem HOT 7 CLOSED

Hi @ankitbioinfo,

I created a branch for your dataloader called wang_dataloader. I already created the basic structure, but some parts need to be adjusted depending on your input.

1. Coarseness of cell type annotation
   

   ncem/ncem/data.py

   Line 2702 in 7bc1b22

   cell_type_merge_dict = {

In your dataset I saw a quite granular annotation for different cell types. We currently recommend to use a cell type annotation that does not specify sub-classes as we inspected that ncem can directly infer spatially dependent sub-states.

The posted link shows a grouping logic as a dictionary to create a coarser annotation. Please feel free to adjust this if you only want to use the data_exploration section of ncem and if you are only interested in analyzing your data and not using the provided model classes.

2. Lateral resolution and spatial omnics technique.
   Could you also provide the lateral resolution of your dataset and the spatial omnics technique of the dataset you provided? Additionally, are the x and y coordinates of your cells given in microns or pixels?

3. Image information (optionally)
   Are your samples aquired from one image/sample or multiple ones?

Thanks!

from ncem.

Hi Anna,

Many thanks for creating the data loader wang_dataloader. I have installed the updated ncem now and when I use the following command

interpreter.get_data(
    data_origin='wang_dataloader',
    data_path=datadir + '/wang/',
    radius=50,
    node_label_space_id='type',
    node_feature_space_id='standard',
)

and I see no data_origin wang_dataloader. I think I am still missing something.

<bound method Estimator.get_data of <ncem.interpretation.interpreter.InterpreterGraph object at 0x7f8dac02ee20>>
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-6-a3613e995f3e> in <module>
      1 interpreter = ncem.interpretation.interpreter.InterpreterGraph()
      2 print(interpreter.get_data)
----> 3 interpreter.get_data(
      4     data_origin='wang_dataloader',
      5     data_path=datadir + '/wang/',

~/Desktop/FabianGNN/wang_ncem/ncem/estimators/base_estimator.py in get_data(self, data_origin, data_path, radius, graph_covar_selection, node_label_space_id, node_feature_space_id, use_covar_node_position, use_covar_node_label, use_covar_graph_covar, domain_type)
    189             graph_covar_selection = []
    190         labels_to_load = graph_covar_selection
--> 191         self._load_data(
    192             data_origin=data_origin,
    193             data_path=data_path,

~/Desktop/FabianGNN/wang_ncem/ncem/estimators/base_estimator.py in _load_data(self, data_origin, data_path, radius, label_selection)
    136             ]
    137         else:
--> 138             raise ValueError(f"data_origin {data_origin} not recognized")
    139 
    140         self.data = DataLoader(data_path, radius=radius, label_selection=label_selection)

ValueError: data_origin wang_dataloader not recognized

from ncem.

Hi @ankitbioinfo,

the dataloader is still in development as we haven't pushed it to the main branch yet.

We first need to clarify the above mentioned three points before I can merge the pull request to release and then with the newest update it will be included.

Could we provide information to the points 1.-3. from my note above? Thank you!

from ncem.

Hi @AnnaChristina

(1) I am not sure whether subclasses of cell types can be inferred because as such no clear marker exist in the gene datset.
If ncem can predict the subclasses of coarser cell types it would be great so whatever you have mentioned in the data.py
it is okay.
(2) The spatial omics technique here is MERFISH and the lateral resolution is 107.9 nm per pixel. The coordinates are in microns.
(3) The samples are from multiple fovs but from one sample.
Thank you.

from ncem.

Hi @ankitbioinfo, thanks for clarifying. Regarding 3): Do you also have a column in your dataset that gives the fov id?

I am asking because we frequently inspected that adding the fov as covariate to ncem can enhance prediction.

from ncem.

Hi @AnnaChristina, I thought FOV ids are not helpful. So I did not mention in nuclei_data_table.xlsx file.
I uploaded the updated one in the following dropbox link. Thanks for helping me to figure it out.
https://www.dropbox.com/scl/fi/thlcj3mfpygfht81lhaez/nuclei_data_table1.xlsx?dl=0&rlkey=6xpyp4a9n5pqxwq2dk3otlzjv

from ncem.

Hi @ankitbioinfo,

we added a dataloader for your dataset. It will be available with the newest release of ncem version 0.3.1

You can already find an initial analysis here.

from ncem.