Hi @saketkc @yuhanH,
I have four datasets with a total of 250K cells that I want to integrate. I would like to use SketchData plus ComBat-seq as integration method, since it returns batch-corrected raw counts.
I am following your sketch integration tutorial and stopped at this sentence:
we emphasize that you can perform integration here using any analysis technique that places cells across datasets into a shared space. This includes CCA Integration, Harmony, and scVI.
Does it mean that all other integration methods either won't work with SketchData or don't meet its assumptions?
Since ComBat-seq returns raw counts, I would still need to use NormalizeData prior to integrated DR and clustering, that is the same normalization used on individual datasets. This is slightly different from your tutorial, where all methods directly work in low-dimensional space and thus just require FindNeighbors + FindClusters after the integration. At the end of the day, however, I will have all cells placed into a shared place, as required, so in principle I guess the workflow may work.
Am I correct?
Hi @saketkc @yuhanH,
I have four datasets with a total of 250K cells that I want to integrate. I would like to use SketchData plus ComBat-seq as integration method, since it returns batch-corrected raw counts.
I am following your sketch integration tutorial and stopped at this sentence:
Does it mean that all other integration methods either won't work with SketchData or don't meet its assumptions?
Since ComBat-seq returns raw counts, I would still need to use
NormalizeDataprior to integrated DR and clustering, that is the same normalization used on individual datasets. This is slightly different from your tutorial, where all methods directly work in low-dimensional space and thus just requireFindNeighbors+FindClustersafter the integration. At the end of the day, however, I will have all cells placed into a shared place, as required, so in principle I guess the workflow may work.Am I correct?