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 framework

  • Fix unintended side-effect of annotation with RCTD on supplied reference anndata

  • Add compatibility of tacco.tools.spectral_clustering() with scipy 1.15

  • Add compatibility of tacco.testing.assert_adata_equals() with np.ndarray objects in .obsm keys

  • Fix sparsity test in tacco.tools._RCTD._round_X() when running tacco.tools.annotate_RCTD(), see #30 thanks to @kmpf

  • Fix #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 taccos tacco.tool.find_regions(). With a clean current installation everything should work again. #1216