workflow queries

[hahahah-single]

Hi ,after Harmony reduction when I run addImputeWeights which reducedDims should I use "IterativeLSI" or "Harmony", and if I should reuse "addClusters" based on IterativeLSI reducedDims to calculate cluster for downstream group coverage generation and call peak, or "Harmony" is also useful.

[rcorces]

I'm not sure if this is a matter of opinion but my feeling is that it is safer to perform addImputeWeights on IterativeLSI reduced dims than on Harmony. However, addClusters should definitely be performed on Harmony and not on IterativeLSI. This isnt perfectly showcased in the manual because we dont really use the Harmony-corrected reduction for much on that dataset.

[plbngl]

Hello thank you for the great work. I have a related question so I am asking it here if that's ok.
I used Harmony to remove sample batch effect and worked well. What would you recommend to run the addGeneIntegrationMatrix() [using an existing seurat object of matched scRNA samples] and addPeak2GeneLinks() functions on: reducedDims = "IterativeLSI" or reducedDims = "Harmony"?
Thank you so much!
Paola

[rcorces]

You should follow the recommendations in the manual.

[plbngl]

Oh my bad, I really did not see that in the manual. Would you mind pointing me to the page where this is specified if that's not a problem ? Thanks!!!

[paupaiz]

Hi Paola, If you created a reducedDims object named “Harmony” then you should pass that one instead of the iterativeLSI object to any functions where you want to account for batch correction.
Ref: https://www.archrproject.com/bookdown/batch-effect-correction-wtih-harmony.html

[YiweiNiu]

Hi @rcorces,

it is safer to perform addImputeWeights on IterativeLSI reduced dims than on Harmony

Could you elaborate a bit on this point? If Harmony has a better batch correction, in theory it has a better neighbor representation for imputation, right? Thanks!