Question on multiple mols2grid instances

[PatWalters]

The callback feature is very useful, thanks! I've been trying to figure out how to implement an enhanced cluster viewer. I've been able to put together an example where I can click on a cluster center in one grid and show the cluster members in a second grid. The part I'm missing is capturing the selections in the cluster member grid. If I click on a cluster in the cluster center grid a second time, the cluster member selections are cleared. Is there a way around this? My example code is in this gist.
https://gist.github.com/PatWalters/ffffa4612e2c143ab7389e539a5c88f7

Also, it would be useful to be able to have a callback in the cluster center grid when checkboxes are not present. However, it appears that when selection=False, clicks in the grid are not recognized.

[cbouy]

Hi Pat, glad you like the new feature!

I had a very quick look at your notebook through Colab and while I could not reproduce this exact issue, something seemingly related happened (a molecule being added to the wrong grid). I think there's a critical race condition at some point, but I'll need to have a better look, probably in two weeks or so.

Also what are you using to retrieve the selections? As of now, each grid have their own independent selection (unless they have the same name, in which case they are overwritten). You can get a specific selection through mols2grid.get_selection(grid_name) which returns a dict, or directly molgrid_object.get_selection() which returns a dataframe.

For the selection=False, that's indeed a bug, I'll try to fix it for the next patch release, thanks for finding it.

[PatWalters]

Thanks for the quick response Cedric! Here's an example. I put slightly modified code on Colab.
https://gist.github.com/PatWalters/4a983edbd5ef18bc9d8f5030b83faed9

Here's how to reproduce the issue.

1. Run the notebook
2. Click on cluster 0 in the cluster centers (top) grid
3. At this point you should see 5 cluster members shown in the bottom grid
4. Click the checkboxes on the first 3 cluster members
5. Click on cluster 1 in the centers (top) grid
6. Click on cluster 0 in the centers (top) grid

Note that none of the cluster members are selected. My expectation is that the selection would persist and the first 3 cluster members would still be selected.

Sorry for the long-winded explanation. I hope this makes sense.

Thanks!

[cbouy]

Hi again Pat and sorry for the delayed response! and thanks for the clear procedure :D

mols2grid generates a plain HTML file and displays it in the notebook when you call mols2grid.display or member_grid.display, so as soon as the callback function is called a new HTML document is created and the state of each checkbox is lost (although it's saved on the python side).

I could try to add a cache_selection keyword (False by default) and every time display is called with cache_selection=True, it will search for a grid with the same name in mols2grid.get_selection() and restore the checkboxes state from there.
Would that fix the problem?

[PatWalters]

Hi Cédric,

That sounds great, thanks!

[cbouy]

Hi again Pat,

I just released the new update (v0.2.0) which includes the cache_selection feature (among others).
I've modified your code to use it:

grid_list = []
for idx,cluster in tqdm(enumerate(cluster_res), total=len(cluster_res)):
    selected_cluster = cluster_res[idx]
    selected_df = df.loc(axis=0)[selected_cluster]
    grid_list.append(mols2grid.MolGrid(selected_df, name=f"members_{idx}", cache_selection=True))

output = widgets.Output()
@output.capture(clear_output=True, wait=True)
def show_cluster_members(data):
    cluster_idx = int(data["Cluster"])
    grid = grid_list[cluster_idx]
    grid._cached_selection = mols2grid.get_selection(f"members_{cluster_idx}")
    view = grid.display(n_rows=3, subset=["img"])
    display(view)

center_grid = mols2grid.MolGrid(center_df, name="centers")
center_view = center_grid.display(n_rows=3,
                    subset=["img", "Cluster"],
                    selection=False,
                    callback=show_cluster_members)
display(center_view)
output

PS: you don't really need to create the grids beforehand anymore since grid creation is now much faster, so you could directly use something like this:

output = widgets.Output()
@output.capture(clear_output=True, wait=True)
def show_cluster_members(data):
    cluster_idx = int(data["Cluster"])
    cluster_members = cluster_res[cluster_idx]
    view = mols2grid.display(df.loc(axis=0)[cluster_members],
                             name=f"members_{cluster_idx}",
                             cache_selection=True,
                             n_rows=3, subset=["img"])
    display(view)

center_grid = mols2grid.MolGrid(center_df, name="centers")
center_view = center_grid.display(n_rows=3,
                    subset=["img", "Cluster"],
                    selection=False,
                    callback=show_cluster_members)
display(center_view)
output

Hope that helps!
Cédric

[PatWalters]

This is awesome, thanks Cedric!