RECEIVED AS A GITHUB ISSUE https://github.com/Oshlack/splatter/issues/16

Up/down regulation of random genes is working a bit weird.

plotPCA(splatSimulateGroups(groupCells=c(1000,1000), nGenes=21), colour_by='Group') gives:

whereas plotPCA(splatSimulateGroups(groupCells=c(1000,1000), nGenes=22), colour_by='Group') leads to:

Is this normal?

OK, this is not related to number of genes directly. Better question is why are these two so different:

plotPCA(splatSimulateGroups(groupCells=c(1000,1000), nGenes=21, seed=4))

plotPCA(splatSimulateGroups(groupCells=c(1000,1000), nGenes=21, seed=5))

By setting nGenes=21 you are creating a simulation with only 21 genes, with the default differential expression probability (0.1) only 2 of these genes will be differentially expressed. Depending on the expression level of these genes and the random fold changes that are simulated PCA may/may not pick up the differences between these groups.

If you repeated the experiment with more genes (nGenes = 10000) it should produce more consistent results. If you want to play around with different numbers of DE genes you can change de.prob.