TACCO 0.5.0 (2026-04-21)¶
Features¶
Refactored the conda_env specification for RCTD, tangram, and SingleR to have more flexible environment specification as well as a fallback to use the current environment.
Implement automatic casting for bisectioning #26
Improve scaling of
tacco.tools.fill_regions()from N**2 to N log(N)Prepared ARM support #24
Fixes¶
Catch edge cases in the anndata2R import functionality (integer count matrices, annotation columns containing NaNs, and categorical indices) relevant for the wrapped R tools
Fix SingleR parameters for SingleR bioconda-version 2.8.0, which corresponds to bioconductor version 3.20.
Restore dtype after
tacco.tools.fill_regions()Arguments of
tacco.tools.fill_regions()are extended and aligned with other functions in the frameworkFix unintended side-effect of annotation with RCTD on supplied reference anndata
Add compatibility of
tacco.tools.spectral_clustering()with scipy 1.15Add compatibility of
tacco.testing.assert_adata_equals()with np.ndarray objects in .obsm keysFix sparsity test in
tacco.tools._RCTD._round_X()when runningtacco.tools.annotate_RCTD(), see #30 thanks to @kmpfFix #31: compatibility with scipy 1.17 issue
Breaking changes¶
Expose RCTD filter parameters n_max_cells, min_UMI, and counts_MIN and deactivate these filters by default to make the filtering transparent on the python level.
Migrate deprecated github action mamba-org/provision-with-micromamba to mamba-org/setup-micromamba. Cleaned up environment.yml in the process, so the old “clean” install using that yml file is not possible anymore._
Miscellaneous¶
Note that numba>=0.62.0 is incompatible with umap-learn<=0.5.12, which influences scanpy’s
sc.pp.neighbors()results and thereby also taccostacco.tool.find_regions(). With a clean current installation everything should work again. #1216