heatmap matrix

[zverozabr]

I am trying to create a heatmap of cluster matrices and their intensity. The data has already been calculated by me. I found in the documentation the possibility of such an approach, but I did not find examples of use to see what the structure of the Zarr archive should be in order to see a similar matrix.

i have this matrix

adata.uns['global_clq']
4 6 7 11 -1 9 0
4 1.854253 0.219517 0.591505 0.347924 0.193056 0.514815 1.158333
6 0.217952 3.031756 0.156767 0.539432 0.313063 0.626126 0.000000
7 0.629680 0.143585 4.425228 0.609649 0.000000 1.300585 0.000000
11 0.408387 0.590003 0.609649 6.237180 0.000000 0.000000 0.000000
-1 0.193056 0.313063 0.000000 0.000000 19.305556 0.000000 0.000000
9 0.450463 0.417417 1.219298 0.000000 0.000000 24.453704 0.000000
0 1.158333 0.000000 0.000000 0.000000 0.000000 0.000000 69.500000
adata.uns['clq_perm']
4 6 7 11 -1 9 0
4 1.000 0.000 0.716 0.131 0.066 0.521 0.808
6 0.000 1.000 0.005 0.550 0.243 0.574 0.000
7 0.806 0.004 0.839 0.641 0.000 0.890 0.000
11 0.218 0.645 0.624 0.762 0.000 0.000 0.000
-1 0.061 0.226 0.000 0.000 0.995 0.000 0.000
9 0.440 0.378 0.804 0.000 0.000 0.598 0.000
0 0.752 0.000 0.000 0.000 0.000 0.000 0.483
Actually, my question is, I would be very happy if you could provide an example of the config you used to create the screenshot in the documentation, or give me the URL of this config, so that I can dissect the Zarr structure myself. I would appreciate your help. I like what I am achieving, but this limitation has slowed me down a bit.

something like this

[ilan-gold]

@zverozabr Mark would know better than I the status of the custom heatmap format, but I don't believe there is any issue except providing a dense zarr matrix, except that its dimensions should match those with which it is coordinated (i.e., points on a UMAP and the obs axis of the heatmap). If you have a matrix from AnnData you should indicate to vitessce that this is what you have (AnnData has a custom sparse on-disk format so perhaps that is your issue). For example from following along here I don't see why you couldn't just point at uns['global_clq'] instead of X as long as it's not coordinated with anything that expects it to be of a certain shape. So something like:

{
  "fileType": "obsFeatureMatrix.anndata.zarr",
  "url": "https://example.com/my_adata.zarr",
  "coordinationValues": {},
  "options": {
    "path": "uns/gloabl_clq"
  }
},

I'm not sure if coordinationValues are required off hand so you may have to provide some dummy ones so that you don't have an issue with your actual obs visualization (since I assume the uns matrix is for clusters, not for obs judging by your labels).

[keller-mark]

The heatmap in Vitessce assumes that one axis represents "observations" and the other represents "features".
If we consider each cluster as an "observation", then in this case we need to kind of trick Vitessce into showing the data.

We need to create a new AnnData object that has the clusters along both obs (observations) and var (features) axes. We can store two matrices aligned to these axes using layers:

from anndata import AnnData
import pandas as pd

obs = pd.DataFrame(index=[4, 6, 7, 11, -1, 9, 0], columns=[], data=[])
var = pd.DataFrame(index=[4, 6, 7, 11, -1, 9, 0], columns=[], data=[])

cluster_adata = AnnData(X=None, obs=obs, var=var)
cluster_adata.layers['global_clq'] = adata.uns['global_clq']
cluster_adata.layers['clq_perm'] = adata.uns['clq_perm']

cluster_adata.write_zarr('clusters.h5ad.zarr')

Then the JSON snippet to configure this would look like:

{
  "fileType": "obsFeatureMatrix.anndata.zarr",
  "url": "https://example.com/clusters.h5ad.zarr",
  "coordinationValues": {
     "obsType": "cluster",
     "featureType": "cluster",
     "featureValueType": "global_clq"
  },
  "options": {
    "path": "layers/global_clq"
  }
},
{
  "fileType": "obsFeatureMatrix.anndata.zarr",
  "url": "https://example.com/clusters.h5ad.zarr",
  "coordinationValues": {
     "obsType": "cluster",
     "featureType": "cluster",
     "featureValueType": "clq_perm"
  },
  "options": {
    "path": "layers/clq_perm"
  }
}

Using the python API it would look like:

from vitessce import (
    VitessceConfig,
    AnnDataWrapper,
    ViewType as vt
)

vc = VitessceConfig(schema_version="1.0.16")

dataset = vc.add_dataset("clusters").add_object(AnnDataWrapper(
    adata_path = 'clusters.h5ad.zarr',
    obs_feature_matrix_path = "layers/global_clq",
    coordination_values={
        "obsType": "cluster",
        "featureType": "cluster",
        "featureValueType": "global_clq"
    }
)).add_object(AnnDataWrapper(
    adata_path = 'clusters.h5ad.zarr',
    obs_feature_matrix_path = "layers/clq_perm",
    coordination_values={
        "obsType": "cluster",
        "featureType": "cluster",
        "featureValueType": "clq_perm"
    }
))

heatmap_global = vc.add_view(vt.HEATMAP, dataset = dataset)
heatmap_local = vc.add_view(vt.HEATMAP, dataset = dataset)

vc.link_views([heatmap_global, heatmap_local], ["obsType", "featureType"], ["cluster", "cluster"])
vc.link_views([heatmap_global], ["featureValueType"], ["global_clq"])
vc.link_views([heatmap_local], ["featureValueType"], ["clq_perm"])

vc.layout(heatmap_global | heatmap_local)