anndataR 1.3.5
- Add a
groupargument toread_h5ad(),write_h5ad(),as_HDF5AnnData()andHDF5AnnData, to read and write anAnnDatastored in a group inside an HDF5 file, e.g. a modality inside a.h5mufile (PR #521, issue #466). - Fix
as_AnnData()failing with`rownames(X)` is not as expectedwhen the assay of aSingleCellExperimentmapped toXhas different dimnames than the object (PR #576, issue #390). - Fix
as_AnnData()dropping the loadings of a Seurat reduction when their row names already match the feature names of the object (PR #592). - Ensure Zarr v2 string arrays don’t get the
vlen-utf8filter twice, since an upcoming version of Rarr might already specify this (PR #557, Huber-group-EMBL/Rarr#111). - Fix
as_Seurat()failing withNo feature overlap between existing object and new layer datawhenvar_namescontain underscores: the underscores are now replaced with dashes in the feature names of theSeuratobject, with a warning (PR #593, issue #476). - Require
rhdf5 >= 2.57.13, which fixes the spuriousNA_character_warning for numeric datasets containingNA, and remove the skips in the file closure tests (PR #579, Huber-group-EMBL/rhdf5#241). - Error when writing a logical matrix containing
NAto H5AD or Zarr, instead of writing anullable-booleanthat neither R nor Python anndata can read back. Convert it to numeric first (issue #545). - Require
Rarr >= 2.1.59, which fixes reading and writing Zarr stores on R 4.6.2: every zstd-compressed array failed there withargument is of length zero(PR #591, Huber-group-EMBL/Rarr#298). - Write a length 1
NA_character_as anullable-string-arrayto Zarr and H5AD, so that it stays missing: Zarr stored it as the string"NA", and Python anndata read it from H5AD as an empty string (PR #585, issue #581). - Regenerate the man pages, which still showed the old default
compressionofwrite_zarr()and missed a contributor (PR #594). - Fix writing empty arrays to Zarr v3, which produced stores that Python anndata couldn’t read: empty dimensions now get a chunk size of 1, as zarr-python does (PR #587, Huber-group-EMBL/Rarr#294).
- Add
backed = TRUEtoread_zarr()andas_ZarrAnnData()to read matrices as ZarrArray-backed DelayedArrays (PR #507). - Add https://bioc.r-universe.dev as a repository in the workflows that check against Bioconductor devel, so that new versions of dependencies can be installed before they reach the Bioconductor repositories (PR #595).
- Add Umami analytics to the {pkgdown} site (PR #590).
anndataR 1.3.4
- Replace the internal
create_zarr()andcreate_zarr_group()helpers withRarr::write_zarr_group()and requireRarr >= 2.1.27(PR #490). - Use zstd by default in
write_zarr()(PR #540). - Write length 1
data.framecolumns and indices as arrays instead of scalars (PR #556, issue #553). - Fix writing integer matrices containing
NAto Zarr, which produced stores that couldn’t be read back: as for H5AD, they are now written as double (PR #546, issue #542). - Keep
NAin double and integer matrices written to H5AD instead of replacing it withNaN, so thatNAandNaNcan still be told apart when reading the file back in R. Python anndata reads both asNaN(PR #546). - Fix the error and warning messages when reading or writing an element fails: the original error is now shown as the cause, instead of being parsed as a cli expression, which failed with
Could not parse cli `{}` expression(PR #547, issue #531). - Check that the indices of sparse matrices fit their shape when reading H5AD files and Zarr stores, with an error that suggests the indices may not be 0-based (PR #547, issue #531).
- Fix writing non-ASCII strings to H5AD files by using UTF-8 for string datasets (PR #555, issue #400).
- Re-enable testing against Python anndata >= 0.13, and remove the warning that
ReticulateAnnDatagave for these versions (PR #569, issue #514). - Skip the Python-written Zarr v2 roundtrip tests when the installed Python anndata can no longer write Zarr v2, as anndata main does (PR #568, issue #514).
- Skip the
Nonekey that Python anndata >= 0.13 uses forXwhen reading layers throughReticulateAnnDataandpy_to_r(), and keepXwhen settinglayers(PR #563, issue #514). - Ignore the
None (.X)entry that Python anndata >= 0.13.2 lists underlayers:when comparing printed objects in the roundtrip tests (PR #564, issue #514). - Drop 3D layers from the layers roundtrip tests, because Python anndata >= 0.13 refuses to write them (PR #571, issue #514).
- Require
Rarr >= 2.1.45to read the sharded Zarr v3 stores that Python anndata >= 0.13 writes by default, because older versions can return wrong values for sharded arrays (PR #572, issue #514).
anndataR 1.3.3
- Add a strict mode for the Python tests: with
ANNDATAR_TEST_REQUIRE_PYTHON=true, the Python skip helpers error with the actual import error instead of skipping, so that a broken Python environment can’t silently skip all Python tests. Used by scverse/integration-testing to test against anndata main (PR #533, issue #515). - Fix writing
data.frames inobsm/varmso that the stored index is theobs_names/var_namesof the parent object instead of R’s automatic row numbers, which produced files that Python anndata refused to read (PR #501, PR #525).
anndataR 1.3.2
- Support the
anndata.AnnDataclass name used by Python anndata >= 0.13 inpy_to_r()andReticulateAnnData, and warn that anndata >= 0.13 is not fully supported yet (PR #510, issue #499). - Add support for
nullable-string-arrayelements in H5AD and Zarr (PR #480). - Fix
hdf5_write_boolean_dataset()writing logical matrices as flat 1D datasets instead of preserving their shape (PR #496). - Remove the workaround for LZF compression of VLen strings and require
rhdf5 >= 2.57.5, which fixes the issue upstream (PR #497). - Pin Python
mudatato< 0.4in theusage_pythonvignette, as mudata 0.4 can no longer read the legacy example.h5mufile (PR #512, issue #511). - Pin Python
mudatato< 0.4in the pkgdown and macOS-Intel check workflows as well, becausepy_require()pins in the vignette are ignored whenRETICULATE_PYTHONis forced (PR #518, issue #511). - Remove the
as.naattribute that rhdf5 2.57.12 adds to datasets containingNAwhen comparing R-written files to Python-written ones in the roundtrip tests (PR #519, issue #517). - Skip the H5AD file-closure tests while rhdf5 2.57.12 emits a spurious
NA_character_warning for numeric datasets containingNA, tracked as a known issue until the fix from Huber-group-EMBL/rhdf5#242 reaches Bioconductor devel (PR #520). - Remove the special handling of Zarr v3 structured datatypes in the tests, now that
Rarrsupports them (PR #498). - Check
RELEASE_*branches against Bioconductor release anddevelagainst Bioconductor devel, instead of checkingdevelagainst both (PR #502). - Run BiocCheck on
RELEASE_*branches as well as ondevel(PR #504).
anndataR 1.3.1
- Add support for reading backed objects using
DelayedArraymatrices, including abackedargument forHDF5AnnDataandInMemoryAnnData, and conversion of backedAnnDataobjects toSingleCellExperiment/Seurat(PR #387). - Add support for writing Zarr v3 stores, using (i)
zarr_formatargument ofwrite_zarr()/as_ZarrAnnData()or (ii) theanndataR.zarr_formatoption. New stores are written as Zarr v3 by default (PR #455). - Write strings as VLen-UTF8 rather than as fixed-width, so that string arrays read back as variable length strings in Python
anndata(PR #455). - Use the format of an existing Zarr store when writing to it and truncate the store when
mode = "w", so that a store can no longer end up with a mix of Zarr v2 and v3 nodes (PR #455). - Enable additional linters and optimise several suboptimal code patterns (PR #453).
- Fix
create_zarr()so that Zarr stores can be created at relative paths and at paths containing regex metacharacters (PR #478). - Fix use of
&instead of&&when validating column names inAbstractAnnData(PR #487). - Update tests to take into account the
Rarrstructured datatype breaking change (PR #462). - Temporarily require Python
anndata < 0.13.0in tests, workflows and the Python usage vignette (PR #481). - Add Zarr to the benchmarks (PR #446).
- Add Bioconductor R-universe checks (PR #470).
- Fix CI runners (PR #457).
- Upgrade to roxygen2 8.0.0.
- Simplify roxygen documentation for the H5AD and Zarr helpers by using
@inheritParams(PR #479). - Update
CITATIONwith the published paper (PR #464). - Mark Zarr support as implemented in the class diagram (PR #460).
anndataR 1.2.3
Backported from the development branch (PR #538):
- Fix writing
data.frames inobsm/varmso that the stored index is theobs_names/var_namesof the parent object instead of R’s automatic row numbers, which produced files that Python anndata refused to read (PR #501, PR #525). - Pin Python to 3.14 in the benchmark and pkgdown workflows, as
h5pyhas no wheels for Python 3.15 yet (PR #534).
anndataR 1.2.2
Backported from the development branch (PR #503, PR #523):
- Add support for
nullable-string-arrayelements in H5AD and Zarr (PR #480). - Fix
hdf5_write_boolean_dataset()writing logical matrices as flat 1D datasets instead of preserving their shape (PR #496, thanks @JuanTejedor!). - Check
RELEASE_*branches against their own Bioconductor version instead of against Bioconductor devel, and run the checks on pushes to release branches (PR #502). - Support the
anndata.AnnDataclass name used by Python anndata >= 0.13 inpy_to_r()andReticulateAnnData, and warn that anndata >= 0.13 is not fully supported yet (PR #510, issue #499). - Pin Python
mudatato< 0.4in theusage_pythonvignette and in the pkgdown and check workflows, as mudata 0.4 can no longer read the legacy example.h5mufile (PR #512, PR #518, issue #511). - Update the Zarr recarray tests to take into account the
Rarrstructured datatype breaking change, so that they pass with both Bioconductor release and devel versions ofRarr(PR #462). - Measure test coverage against the Bioconductor version the branch targets instead of always against devel (PR #523).
- Drop the R-devel setup workaround for the Windows and macOS ARM check jobs, which installed R devel against Bioconductor 3.23 packages (PR #523).
anndataR 1.2.1
Backported from the development branch (PR #489):
- Fix
create_zarr()so that Zarr stores can be created at relative paths and at paths containing regex metacharacters (PR #478). - Fix conditional expression in
AbstractAnnDatamatrix validation (PR #487). - Require Python
anndata<0.13.0in tests, workflows and vignettes (PR #481). - Update
CITATIONwith the published paper (PR #464).
anndataR 1.1.1
- Fix CI (PR #418).
- Add continuous benchmarking using bencher (PR #423, PR #425).
- Handle unnamed
SingleCellExperimentassays inas_AnnData()by automatically assigning names with a warning (PR #420). - Optimise sparse matrix reading performance by avoiding
Matrix::sparseMatrixand constructing objects manually (PR #417). - Allow manually setting chunk size for HDF5 writes (PR #424).
- Add auto-chunking for HDF5 writes to improve performance (PR #424).
anndataR 1.1.0
- Bioconductor 3.23 devel
- Separate reading and writing keys in
HDF5AnnData(PR #392). - Add a check for 0 dimensions when converting to
Seurat(PR #393). - Warn that matrix dim names will not be written to H5AD files (PR #396).
- Add a warning to
.from_SCE_convert()(PR #402). - Warn when a rotation is provided but not stored in a
LinearEmbeddingMatrix(PR #407). - Fix linting issues (PR #389).
- Add Bioconductor badges (PR #375).
- Update {pkgdown} rendering (PR #379).
- Force loading the environment in the Python vignette (PR #394).
- Add issue templates (PR #395).
- Update GitHub Actions based on those used by {rhdf5} (PR #378).
- Set different cache versions in GitHub Actions (PR #403).
- Update CI (PR #416).
anndataR 1.0.2
- Fix broken links (PR #430).
anndataR 0.99.4
- Address minor issues from Bioconductor checks (PR #371)
- Combine the
opts_chunks$set()calls in the Python usage vignette - Replace
\donttrunwith\donttestin man pages
- Combine the
anndataR 0.99.3
New functionality
- Implemented an
AnnDataViewclass, which provides a lazy view of anAnnDataobject without copying data (PR #324) - Implemented S3 methods for
AbstractAnnDataobjects:dim,nrow,ncol,dimnames,rownames,colnames, and[(PR #324) - Add
ReticulateAnnDataclass for seamless Python integration via {reticulate} (PR #322) - Add
AGENTS.mdwith instructions for developers (PR #367).
Major changes
- Refactor obs/var_names handling for improved data consistency (PR #328)
-
InMemoryAnnDatanow storesobs_namesandvar_namesas separate private fields instead of relying on rownames ofobs/vardata.frames -
HDF5AnnDatamaintains separate obs/var names management to ensure consistency between obs/var data.frames and dimnames - All matrix data (
X,layers,obsm,varm,obsp,varp) is now stored internally without dimnames for consistency - Dimnames are added on-the-fly when users access data, ensuring proper obs/var name display
-
Minor changes
- Refactor setter methods in
HDF5AnnDataandInMemoryAnnDatato use pipe operators for cleaner code (PR #328) - Add explanatory comments for matrix generation alignment with Python dummy-anndata (PR #328)
- Add
get_generator_types()function return allowed/example types forgenerate_dataset()(PR #354) - Add checks for type arguments to
generate_dataset()(PR #354) - Generalise the layers created by
generate_dataset()whenformat = "Seurat"(PR #354)
Bug fixes
- Add compression parameter to additional write operations in
HDF5AnnDatafor consistency (PR #328) - Directly use
obs_namesandvar_namesproperties instead of corresponding indirect S3 methodsrownamesandcolnames(PR #328) - Fix error message variable name in
.validate_aligned_array()method (expected_colnames → expected_rownames) (PR #328) - Fix Seurat conversion for PCA loadings with variable feature subsets (PR #328)
- Seurat PCA loadings only contain variable features, not all genes
- Now properly expands loadings matrix to include all genes with zeros for non-variable features
- Adds warning when rownames don’t match var_names during conversion
- Avoid writing character datasets to H5AD files with LZF compression as it causes R to crash (PR #356)
- Handle slots that may have incomplete dimensions when converting from
Seurat. These are now skipped with a warning instead of indirectly raising an error. (PR #369)
anndataR 0.99.2
- Add
@returnto R6 object man pages to address {BiocCheck} warning (PR #319)
anndataR 0.99.1
- Resubmit because of failing Bioc checks (PR #318)
anndataR 0.2.0
Breaking changes
- Switch the HDF5 back end to use the {rhdf5} package instead of {hdf5r} (PR #283, Fixes #272, #175, #299)
- This addresses various issues related to H5AD files and allows better integration with Bioconductor. Most of the previous known issues have now been resolved.
- It also greatly improves compatibility with H5AD files written by Python anndata
- NOTE: Make sure to install {rhdf5} instead of {hdf5r} to be able to read and write H5AD files!
Major changes
- Updates for compatibility with Python anndata >= 0.12.0 (PR #305, Fixes #304)
- Add helpers for reading/writing
NULLvalues to/from H5AD files - Writing of
NULLvalues can be disabled by settingoption(anndataR.write_null = FALSE)to allow the files to be read by Python anndata < 0.12.0
- Add helpers for reading/writing
- A
countsordatalayer is no longer required duringSeuratconversion (PR #284)- There will still be a warning if neither of this is present as it may affect compatibility with {Seurat} functions
anndataR 0.1.0 (initial release candidate)
Initial release candidate of {anndataR} including:
- Native reading and writing of H5AD files
- R implementations of
InMemoryAnnDataandHDF5AnnDataobjects - Conversion between
AnnDataandSingleCellExperimentorSeuratobjects - Extensive function documentation and vignettes demonstrating usage
- Comprehensive unit testing and identification of known issues