Updates

[orenbenkiki]

This issue will be updated every time we create a new release or have other important updates that impact metacells users. If you are using the python metacells package, you are encouraged to click on the "subscribe" button to the right, and github will update you every time we post an update here. This requires you to have a github account (which is free and easy to create).

[orenbenkiki]

New release 0.9.3

TL;DR: Users are encouraged to pip install --upgrade metacells to obtain this newer version as this should produce better-sized metacells by default. If needed, it also allows for better manual control over the metacell sizes.

Compatibility:

This API of this version is backward compatible with 0.9.[0-2], and is extended to allow specifying a new parameter target_metacell_umis (default of 160,000) in addition to the existing target_metacell_size in cells (whose default reduced from 96 to 48).

The metacells computed by this version will be different from these computed by versions 0.9.[0-2].

Details:

This version solves the issue that previous 0.9.[0-2] versions metacell sizes were too sensitive to the number of UMIs in the cells, which created too-large metacells in most cases.

It is always advisable to look at the distribution of UMIs per metacell and cells per metacell (in the QC page in MCView) and consider adjusting one of the targets to better fit your data.

[orenbenkiki]

Published version 0.9.4 with minor fixes, mainly adapting the package to updated versions of numpy and scipy which have subtly changed some behaviors and/or deprecated some functionality. Results "should" be identical to 0.9.3 (given you use the same version of numpy and scipy).

[orenbenkiki]

Published version 0.9.5 with bug fixes. The major one is a avoiding a potential deadlock due to threadpoolctl not being thread safe (which you'd expect with the word "thread" right there in the name ... sigh), plus some other very minor fixes. Results "should" be identical to 0.9.4 and 0.9.3 (given you use the same version of numpy and scipy).