Current state of hashing summary I have added all methods for which we already have nf-core modules to our hashing summary script (

htodemux

,

multiseq

,

demuxem

,

hasheddrops

and

gmmdemux

). We are currently still missing

bff

(PR) and

hashsolo

for hashing. @Nico Trummer, should I continue without these modules for now? Or do we have a solution for

bff

and

hashsolo

?

For hashsolo the module exists here and for BFF there is the PR which fails due to too high memory consumption Can be fixed by changing the test data or checking if the tool has parameters that can lead to reduced memory footprint

Thanks! Then I'll add hashsolo.

You cannot directly see it, but in some previous commit I increased the memory limit to 6GB and then it worked You could also try 5GB, 4.5 GB etc. until you know (debugging with the CI, wohoo xD) Just need to make sure we don't merge any alterations to the global memory limit, test-wise it's fine

Got it! So, for testing purposes, I gradually decrease the memory limit. Then I can estimate by how much we exceeded the memory limit. But then I still have to adjust the parameters, or in the worst case, the test dataset to stay below the 4 GB limit, right? Wohooo, CI debugging 🥳

I guess Seo can say more about this, but I am pretty sure you can just change it

Thanks to @Nico Trummer for your tip on how to debug the memory problem. I learned a lot again.

Bff

is now working. party dino I would be happy if you could review it as soon as you have time (PR).

Done

I opened an issue for hashsolo because all barcodes are negative for our testing data. They have already helped me, but let me know if you have any ideas about the problem. I am just beginning to gain experience working with this kind of data.

I can't really help you with these things

No problem. We could solve it already. ✅ Thank you @Nico Trummer!

I executed the tests locally on my laptop and get the same hashes as the ones that are created during the CI tests This means you can find the differences by comparing the files generated on github codespaces with the ones generated on my computer. Once you know what's different, you can either soften the tests or update the template script I will send you the

test_hashsolo.csv

which is generated as part of the

hto_data_10x

test, then you can compare

Thank you @Nico Trummer! It is a rounding problem:

diff test_hashsolo.csv test_hashsolo-2.csv
2c2
< ACTATCTAGATGCCTT,35.0,314.0,1.0,0.0,0.0008512746193954035,0.9982658741923409,0.0008828511882637065,MS-12
---
> ACTATCTAGATGCCTT,35.0,314.0,1.0,0.0,0.0008512746193954033,0.9982658741923409,0.0008828511882637064,MS-12
12c12

How many digits do you recommend rounding to? Deseq2 uses

round_digits = NULL

by default. So do I just need to change the

round_digits

with

ext.args

to test the module? Or should I always round up a little by default for this module?

Not rounding by default is fine, people who value reproducibility so much that this matters can use

ext.args

to activate the rounding And for testing you can check at which number of decimals it becomes stable, I would probably start with 10 based on the example values you sent (if it does not work, you can still decrease)

Thanks! I now pass all tests in the CI. I would be happy if you could review it as soon as you have time (PR).