plotDensities fails if logcounts is not of class numeric

[michellemeier27]

Hi,

I've been playing around with your package and noticed that the plotDensities function failed on my SCE object because the logcounts assay is in dgCMatrix format instead of class numeric matrix.

I usually use Seurat, so I convert my seurat object like this:
sceObject <- as.SingleCellExperiment(seurat.object)

The getting everything ready for Cepo:
log2counts <- logcounts(sceObject) %>% as.matrix()
cellType <- sceObject$L1_Seurat_cluster_predicted_phenotype %>% as.vector()

Followed by running Cepo:
ds_res = Cepo(exprsMat = log2counts, cellType = cellType, prefilter_pzero = 0.7, computePvalue = 200, minCells = 10 )

Then finally, plotting:
plotDensities(x = sceObject, cepoOutput = ds_res, genes = 'lyve1b', assay = "logcounts", celltypeColumn = "L1_Seurat_cluster_predicted_phenotype")
Which gives me the following error:
Error in fix.by(by.x, x) : 'by' must specify a uniquely valid column

However, I can work around that by fixing the logcounts in sceObject:

sceObject_2 <-sceObject
log2counts <- logcounts(sceObject) %>% as.matrix()
class(log2counts) <- "numeric"
logcounts(sceObject_2) <- log2counts

And then plotDensities works fine.

I think when you subset the expression matrix on line 126, it results in a format that is then not compatible with line 134 reshape2::melt. Not sure if this is something you'd consider a bug, but as the community seems to move towards using sparse matrices whenever possible it could be worth making it compatible with them.

Cheers,
Michelle