Hello everyone, I am working on scRNA-Seq data analysis. I mainly focused on Scater and Seurat. For some visual properties I (and my PI) want to use Seurat too but it only has LogNormalization. I believe it just takes the logarithm (in Seurat it is natural log not log2 I think).

I want to use deconvolution method which is provided by Scater package. Convert() function of Seurat transforms a SingleCellExperiment to Seurat Object but I think I causes the loss of some metadata.

I used Seurat until normalisation and converted it to SingleCellExperiment object, normalised it (without transforming values to log). What should I do now? I can convert the SingleCellExperiment object to Seurat object, log normalize it and keep going as in tutorial (I am not sure how scaling factor effects). As another option, I can change only data matrix (seuratdata@data and exprs/normcounts(singlecellobject).

Is there a simple way to do it?

Thank you all.

If you're talking about computeSumFactors; that function lives in scran, not scater. In addition, you should be supplying the raw counts (not scaled, not transformed) to this function, see the various examples at https://bioconductor.org/packages/simpleSingleCell.

So if you're starting from the raw data anyway, you might as well start from an SCE, do all the necessary normalization, and convert this to a Seurat object. You can inspect the internal Seurat slots to confirm if the conversion preserved the pre-existing normalization. If it doesn't... I'm not sure if there's much that you can do. Fiddling directly with the slots may be the only option but is very dangerous because you're bypassing input checks that would otherwise guarantee the validity of the object and its contents.