Where are HE_registered and HE_nuc_registered?

Competition (ended) Broad Institute Crunch #1
1

I’m sorry for the next dumb question, but where are HE_registered and HE_nuc_registered? In the description, it says they are in the Zarr files, but there only HE_original and HE_nuc original are present.

In the data that I downloaded, I cannot locate them anywhere.

Further, what I also found confusing is this line in the basic-EDA notebook:
sdata.pl.render_images(“HE_nuc_original”).pl.show(ax=axes[0], title=“H&E registered image”, coordinate_systems=“global”)

Why is the HE_nuc_original loaded as the registered image?

Thank you for any help!

Best,
Thomas

2

I couldn’t find them neither. For all zarr files I downloaded, they are all like

SpatialData object, with associated Zarr store:  somewhere
├── Images
│     ├── 'HE_nuc_original': DataArray[cyx] (1, 17000, 20992)
│     └── 'HE_original': DataArray[cyx] (3, 17000, 20992)
└── Tables
      ├── 'anucleus': AnnData (144704, 460)
      └── 'cell_id-group': AnnData (277046, 0)
with coordinate systems:
    ▸ 'global', with elements:
        HE_nuc_original (Images), HE_original (Images)
3

Ok, a colleage of mine just found out that the registered images are not provided for crunch 1. That is why they are not there. However, the question is now if this means that we have to do this coordinate matching ourselfes if we want to have a meaningful prediction.

image
image662×113 3.13 KB

4

I found some coordinates in the anucleus AnnData.obsm but they are mismatching with the HE original imgs. Beside some mirroring and rescaling, there are probably some cropping operations done as well. I assume those coordinates are corresponding to some preprocessed imgs instead of the originals. Please correct me if I misunderstood anything.

5

You need to use the image HE_nuc_original (H&E nucleus segmentation on original image) to find the positions on the HE_original image; see save_cropped_cells in the quickstarter basic-eda.
You can also directly use the cell_id obtained via regionprops(sdata['HE_nuc_original'][0, :, :].to_numpy())

1 Like
6

The other images described in the full specification will be available in the large data version coming in the next few days.

1 Like
7

Hi! Could you let me know how I can download the large data version?

8

on the crunch setup get token webpage, there is option for data size

image
image434×253 14 KB

2 Likes
9

Hello,

Are there still ways to download the large data version? I lost my data for some reason. Could someone please help?

Thanks!

1 Like
10

Yes

If you do not have an environment setup:

  1. Go on: https://hub.crunchdao.com/competitions/broad-1/submit/cli
  2. Get a setup command
  3. Append --size large at the end
  4. Execute the command

If you already have an environment setup:

  1. Just do crunch download
1 Like
11

Thanks a lot, Enzo.

But looks like I can’t create a new model anymore.

Even though the new model won’t be eligible for joining the competition, can I be allowed to create a new model just to experiment with it?

Appreciate it.

1 Like
12

I created you a model.
But normally we don’t allow model creation after the competition is over to not inflate our statistics.

1 Like
13

Thank you so much Enzo!

In this case, I won’t submit the score of the new model so no worries about the inflation of the statistics.

Appreciate your kindness.

2 Likes