May I ask if SC3 is suitable for raw count matrix generated by Dropseq/Indrop? Usually the output of them are not FPKM count matrix as suggested in the package manual.

Also, I got an issue with SingleCellExperiment issue. When I use the function normcounts(sce), an error shows up like

Error in assay(object, i = exprs_values) :
  'assay(<SingleCellExperiment>, i="character", ...)' invalid subscript 'i'
'i' not in names(assays(<SingleCellExperiment>))

or if I calculate CPM here as  calculateCPM(sce, use.size.factors = FALSE)

Error in (function (classes, fdef, mtable)  :

  unable to find an inherited method for function ‘featureNames’ for signature ‘"SingleCellExperiment"’

Both of the mistakes are shown up even if use the example data embedded in the scater package as guided by the tutorial (https://www.bioconductor.org/packages/devel/bioc/vignettes/scater/inst/doc/vignette-intro.html)

Based on the mistakes on scater/singlecellexperiment set, I just could not start with SC3...

For the SingleCellExperiment side: the error message is telling you that it can't find the relevant assay. This means that your SingleCellExperiment doesn't have a matrix named normcounts in the assays slot, hence normcounts(sce) fails. Admittedly the error message could be more informative, though any suggestions in this regard should be directed to the SummarizedExperiment maintainers.

Regarding the featureNames error; I don't know where that comes from. Make sure you're using the latest version of scater (1.6.something). And when asking questions, it is usually proper to post a minimum working example and your session information, see http://www.bioconductor.org/help/support/posting-guide/.

In addition to Aaron's answer, please note that SC3 operates on logcounts slot of the SingleCellExperiment object. So if you would like to cluster raw counts you need to write them directly to the logcounts slot when creating a SingleCellExperiment object (this step is also shown in the SC3 vignette). But I recommend to normalise and log-transform your data before doing clustering.