• First published version.


  • More robust graph partition.

  • Allow forcing feature genes.

  • Rename “project” to “convey” to prepare for addition of atlas projection functionality.


  • Switch to new project template.

  • Fix some edge cases in the pipeline.

  • Switch to using psutil for detecting system resources.

  • Fix binary wheel issues.

  • Give up on using travis-ci.


  • Add inner fold factor computation for metacells quality control.

  • Add deviant fold factor computation for metacells quality control.

  • Add projection of query data onto an atlas.

  • Self-adjusting pile sizes.

  • Add convenience function for computing data for MCView.

  • Better control over filtering using absolute fold factors.

  • Fix edge case in computing noisy lonely genes.

  • Additional outliers certificates.

  • Stricter deviants detection policy


  • Improved and published projection algorithm.

  • Restrict number of rare gene candidates.

  • Tighter control over metacells size and internal quality.

  • Improved divide-and-conquer strategy.

  • Base deviants (outliers) on gaps between cells.

  • Terminology changes (see the README for details).

  • Projection!


  • Fix build for python 3.11.

  • More robust gene selection, KNN graph creation, and metacells partition.

  • More thorough binary wheels.


  • Fix numpy compatibility issue.

  • Fix K of UMAP skeleton KNN graph (broken in 0.9.1).


  • Allow specifying both target UMIs and target size (in cells) for the metacells, and adaptively try to satisfy both. This should produce better-sized metacells “out of the box” compared to 0.9.[0-2].


  • Fix minor bug in regularization of metacell fractions.

  • Fix issue with canonical sparse matrices after downsampling (probably due to scipy.sparse updates?)

  • Fix using deprecated AnnData APIs.