Problem introduced by 0.7.5.1

[orenbenkiki]

I have an AnnData object which contains a sparse matrix in a layer.
This is accessible from R in anndata 0.7.5, but the data is returned as NULL in 0.7.5.1.
I can replicate the issue by switching between the two installed versions of anndata w/o changing anything else in the environment.

[rcannood]

Thanks for bringing this to my attention. Can you give me a reproducible example? :)

…

On Mon, 22 Mar 2021, 13:29 Oren Ben-Kiki, ***@***.***> wrote: I have an AnnData object which contains a sparse matrix in a layer. This is accessible from R in anndata 0.7.5, but the data is returned as NULL in 0.7.5.1. I can replicate the issue by switching between the two installed versions of anndata w/o changing anything else in the environment. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#6>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEHFKWCKTRPJOEPTEN33UTTE4Z3LANCNFSM4ZTAW6BA> .

[orenbenkiki]

My sample file is >20M compressed (>50M uncompressed). Don't know if the size has anything to do with the issue. Since it is >10M, I can't attach it here. I uploaded it to Google Drive:

https://drive.google.com/file/d/137p_KZ83NuG8hTP8SFMj5HgIQS5eOXhS/view?usp=sharing

The data in question is adata$layers$downsampled. It is a sparse matrix in 0.7.5 and NULL in 0.7.5.1.

Let me know if that works and when I can delete it from there.

[rcannood]

Aha, I see! Thanks for the file, I managed to download it.

The issue is that in anndata 0.7.5, the anndata._core.aligned_mapping.Layers Python classwas being transformed into a list, which somewhat limited its usability. In anndata 0.7.5.1 I made the switch to having a separate R6 class for the Layers Python class, causing the aforementioned issue.

One solution would be to simply use adata$layers[["downsampled"]] instead.

I'm considering dynamically creating new active fields for each of the layers inside of the anndata object, so you can use the anndata$layers$downsampled again. This may or may not be a good idea. Let me think about it for a bit :)

[orenbenkiki]

Hmmm, didn't occur to me to try [['downsampled']] instead of $downsampled. That's a reasonable workaround, although I'll miss the simpler form.