Add H5AD read/write helpers

[lazappi]

Add user functions for reading/writing H5AD files to/from SingleCellExperiment/Seurat objects. This will likely be the main interface for most normal users.

Also adds functions for creating dummy data objects (based on the dummy_data() test helper).

[lazappi]

@rcannood The CI here (and on other PRs) is failing on all platforms with an error that I think is coming from your {anndata} package (in the round-trip tests). I can't replicate it locally, though to work out what is going on. Any thoughts on how you want to handle this?

[rcannood]

Sigh! That issue has been popping up everywhere. It's because scipy 1.11 introduced some breaking changes. The development version of reticulate (at rstudio/reticulate) has resolved the issue, but the changes still need to be released on CRAN. I'll add a workaround to the repo for now.

[rcannood]

@lazappi The main branch contains a workaround by switching to the development version of reticulate for now. Can you merge the main branch into this branch?

[rcannood]

These changes make sense to me, though I'd like to allow users to also return hdf5anndata objects instead of only Seurat or SCE. Is this ok with you?

[lazappi]

These changes make sense to me, though I'd like to allow users to also return hdf5anndata objects instead of only Seurat or SCE. Is this ok with you?

I would prefer not to do this. Part of the reason for having these wrapper functions is to hide the HDF5AnnData interface object from most users. I can't see many use cases for interacting with it directly and I feel like anybody "advanced" enough to want to use it should be able to create the objects manually.

[rcannood]

Personally, I see no downsides to enabling this functionality, and again it's something that I actually need in order to use the package myself. Sure, I could do it the roundabout way without using read_h5ad, but I don't know why this should be the case when we might as well just add this functionality.

However, I can agree to not making the default. Would this be okay with you?

[lazappi]

I guess I can be convinced but I would maybe want to see what other people think. I think we just have a different philosophy/use case in mind. I know you have a use for it and could use it safely but you aren't a normal user and I worry about the average person misusing the interface objects if it's too easy to create them. Maybe I'm just paranoid though.

[rcannood]

I merged some changes into this PR.

@lazappi Do you have an idea why the devel branch fails when reading the obs in certain unit tests?

[rcannood]

Namely:

 ── Error ('test-HDF5-read.R:127:3'): reading H5AD as Seurat works ─────────���────
<subscriptOutOfBoundsError/error/condition>
Error in `element[["mask"]]`: subscript out of bounds
Backtrace:
    ▆
 1. ├─anndataR::read_h5ad(file, to = "Seurat") at test-HDF5-read.R:127:2
 2. │ └─anndataR::to_Seurat(adata) at anndataR/R/HDF5-read.R:452:2
 3. │   └─base::ncol(obj$obs) at anndataR/R/Seurat.R:42:2
 4. └─anndataR (local) `<fn>`() at anndataR/R/Seurat.R:42:2
 5.   └─anndataR:::read_h5ad_element(private$.h5obj, "/obs", include_index = FALSE) at anndataR/R/HDF5AnnData.R:43:8
 6.     └─anndataR (local) read_fun(file = file, name = name, version = version, ...) at anndataR/R/HDF5-read.R:67:2
 7.       └─anndataR:::read_h5ad_collection(file, name, column_order) at anndataR/R/HDF5-read.R:361:2
 8.         └─anndataR:::read_h5ad_element(...) at anndataR/R/HDF5-read.R:416:4
 9.           └─anndataR (local) read_fun(file = file, name = name, version = version, ...) at anndataR/R/HDF5-read.R:67:2
── Error ('test-HDF5AnnData.R:32:3'): reading obs works ────────────────────────
<subscriptOutOfBoundsError/error/condition>
Error in `element[["mask"]]`: subscript out of bounds
Backtrace:
    ▆
 1. └─anndataR (local) `<fn>`() at test-HDF5AnnData.R:32:2
 2.   └─anndataR:::read_h5ad_element(private$.h5obj, "/obs", include_index = FALSE) at anndataR/R/HDF5AnnData.R:43:8
 3.     └─anndataR (local) read_fun(file = file, name = name, version = version, ...) at anndataR/R/HDF5-read.R:67:2
 4.       └─anndataR:::read_h5ad_collection(file, name, column_order) at anndataR/R/HDF5-read.R:361:2
 5.         └─anndataR:::read_h5ad_element(...) at anndataR/R/HDF5-read.R:416:4
 6.           └─anndataR (local) read_fun(file = file, name = name, version = version, ...) at anndataR/R/HDF5-read.R:67:2

[lazappi]

I think this might have to do with the changes to {rhdf5}. The nullable arrays are now read natively and you don't have to mess around with the masks yourself. I've fixed this in {zellkonverter} already so I can port that other here if you like.

[rcannood]

I see! Is this related to #83 ?

[lazappi]

Yes. Although I'm not sure if we are testing against the release or devel versions of Bioconductor packages (this change is only in devel).

[rcannood]

The devel R CMD check uses the upcoming release of R, so it probably also has to use the development version of Bioconductor.

I created a PR (#90) which merges #83 into #88, which seems to have solved the issue. @lazappi Is it ok to merge #90?

[lazappi]

I made one minor comment but otherwise yes.

[rcannood]

Thanks! #90 was merged. Can this PR also be merged?

[lazappi]

Yeah, I think so