search

markowetzlab / scDNAseq-workflow

calling absolute copy numbers on single cell DNA sequencing data - stagging repo for public release
4 stars 2 forks source link

scUnique segmentation sensitivity #25

Closed annahyan closed 6 months ago

annahyan commented 6 months ago

Hi,

I managed to run scUnique but the segmentation seems too strict for our data. Is there a parameter I could play with akin to gamma in the copynumber package to fine-tune the granularity?

Thanks, Anna

leachim commented 6 months ago

You can change the sensitivity of the initial scAbsolute segmentation (via the change_prob variable). It is a very conservative segmentation by default (change_prob=1e-3) and so you could try to increase it to 1e-2 or higher to pick up more copy number changes. This can be changed in the worflow/scripts/run_scAbsolute.R script For scUnique, the variable to change is again change_prob; it is encoded as part of the script calling, but you can just remove the if else statement in line 133-135 of the run_scUnique.R script, and hard code it to a value of your liking. Again, values closer to 1 should be more sensitive, so I would suggest increasing it to 0.1 or 0.2 There is a graph in the scUnique publication that we will publish soon with a precision recall curve for this parameter.

leachim commented 6 months ago

Maybe as a side comment, if you change it in scAbsolute the runtime will increase quite a bit, because you will get many more segments and MEDICC2 is relatively slow. So I would first try to change the scUnique variable, unless you believe that the copy number trees are really wrong, then you can try to play around with the scAbsolute parameter.

annahyan commented 6 months ago

Thanks! I'll play around with the parameters!