VDJ usage as Sankey plot [REPLACEMENT ISSUE] about scirpy

[grst]

In GitLab by @szabogtamas on Feb 15, 2020, 16:59

I have fixed the code for VDJ plots I had in the previous attempt of the TCR pipline.
First we have to finish tidying up the current merge requests, but I would propose something like the function at the very end of this notebook.
VDJ_usage.ipynb

[grst]

In GitLab by @grst on Feb 18, 2020, 14:44

That looks pretty cool :)

[grst]

In GitLab by @szabogtamas on Feb 26, 2020, 17:44

mentioned in merge request !14

[grst]

In GitLab by @grst on Mar 16, 2020, 15:08

changed title from VDJ usage as {-a Sankey plot?-} to VDJ usage as {+Sankey plot+}

[grst]

In GitLab by @szabogtamas on Mar 25, 2020, 14:19

Ok, so the current implementation was dead slow. The cause is funny: there is a line that groups by the five VDJ loci and then aggregates sum of cell weights (say, cell numbers). Pandas solves this task by creating a row for every possible VDJ combinations (got more than 4 million rows for 30 cells) and the the next step, joining the loci in a single column is applied on all these rows instead of the less than 30 that actually has a value. Takes more than 10 minutes for a simple test.
Taken this inefficiency of the built-in pandas function, I will go with a more verbose, but hopefully faster approach using the valid combinations only.

[grst]

In GitLab by @szabogtamas on Mar 25, 2020, 19:57

mentioned in merge request !25

[grst]

In GitLab by @grst on Mar 28, 2020, 12:24

closed via merge request !25

[grst]

In GitLab by @grst on Mar 28, 2020, 12:24

mentioned in commit 4b76f27