0.1.9

Hotfixes

• Removed sklearn dependency (#73)

0.1.8

New features

• Added a verbose parameter to the MCMC sampling methods to toggle the printing of progress bar, etc.

Bugfixes

• Fixed a bug in util.datavisualization.boxplots (#55)
• updated the readthedocs settings, which failed to build

0.1.7

Update to latest tensorflow and tensorflow-probability versions. No changes to any funtionality
Changelog:

• Updated scCODA to be compatible with the latest versions of tensorflow (2.8) and tensorflow-probability (0.16.0)
• Updated and extended the unit tests
• Got rid of some frequent warnings:
  -> mean of empty slice when calculating the posterior effects of the reference cell type
  -> transforming to str index when importing data
  -> future warnings for pd.append
  -> future warning for str.replace in the tutorials

0.1.6

Publication release! scCODA has been published at Nature Communications.
Release notes:

• Bugfix: Order of samples when importing data directly from scanpy
• Small adjustments to the documentation

0.1.5

Small release with minor enhancements. This also serves as the release for paper publication!
Changelog:

• HMC with DA and NUTS sampling are now supported officially and do not give out warning messages anymore
• The implementation of ANCOM-BC now has an adjustable FDR parameter alpha
• It is now possible to select a subset of cell types in sccoda.util.data_visualization.boxplots
• Some typo fixes in the documentation

v0.1.4

New features

• scCODA now supports tensorflow 2.4+ and tensorflow-probability 0.12+
• Using HMC sampling with dual-averaging step size adaptation (sample_hmc_da()) and No-U-Turn sampling (sample_nuts()) is no longer discouraged
• Added a progress bar to all sampling methods

Enhancemants

• ANCOM-BC has now an adjustable FDR parameter alpha

Bugfixes

• Fixed a bug in util.result_classes.credible_effects, which still used the old selection method from version 0.1.2

v0.1.3

New features

• Revised hierarchical model formulation. For more info, please refer to the latest revision of the paper
• Added FDR control to scCODA. This changes the way credible results are calculated. The FDR level can be adjusted after inference via result.set_fdr
• Added ANCOM-BC model for comparison
• Added Beta-Binomial model via corncob for comparison

Enhancemants

• Renamed model.dirichlet_models to model.scCODA_model
• Extended data generation function to generate multiple-effect data
• Model evaluation in models.other_models can now deal with multiple effects and different numbers of cell types
• Added level_order parameter to viz.boxplots

Bugfixes

• Fixed a bug in viz.rel_abundance_dispersion_plot that displayed cell type absence instead of presence
• Adjusted zero imputation to add pseudocounts only to zero entries

Documentation and tutorials

• Added a section about FDR control to the "getting started" tutorial

0.1.2.post1

New features

• Added automatic reference selection. Using sccoda.util.comp_ana.CompositionalAnalysis(..., reference_cell_type="automatic") selects a suitable reference cell type. The cell type selected is the one with the least dispersion in relative abundance that is present in at least 90% of samples.
• Added sccoda.util.data_visualization.rel_abundance_dispersion_plot to visualise the automatic reference selection process.
• Added a level_order parameter to sccoda.util.data_visualization.stacked_barplot that allows to change the order of bars. See #19

Enhancements

• Streamlined the models in sccoda.model.other_models to use the same functions

Bugfixes

• Fixed a bug when importing the ancom model from skbio

Documentation and tutorials

• Added a tutorial on how to use sccoda.model.other_models
• Added a tutorial section for sccoda.util.data_visualization.rel_abundance_dispersion_plot
• Added an advanced tutorial section on selecting credibly changing cell types via cycling over all possible references

0.1.1.post1

Bugfixes

• Limited versions of tensorflow (2.3.2) and tensorflow-probability (0.11)