scRNA-seq : runPCA() and denoisePCA()
1
0
Entering edit mode
Bogdan ▴ 660
@bogdan-2367
Last seen 2 hours ago
Palo Alto, CA, USA

Dear Aaron, and all,

referring to the tutorials on scRNA-seq :https://bioconductor.org/packages/release/workflows/vignettes/simpleSingleCell/inst/doc/reads.html,

if I may add another question about runPCA() and denoisePCA() :

i can see that after the step "Modelling the technical noise in gene expression", we use denoisePCA() function to remove late PC that may account for technical noise.

however, i do not see the function runPCA() as part of the computational workflow, and I dunno at which step the PCA is performed .. Or, does denoisePCA() also include runPCA() ?

thank you very much again,

-- bogdan

ps1 : if I may include another question : shall the output of plotReducedDim() and plotPCA() look the same ? Thanks a lot !

ps2 : also -- does plotTSNE() need runTSNE() in a previous step ? Thank you !

scater scran • 1.1k views
3
Entering edit mode
Aaron Lun ★ 28k
@alun
Last seen 15 hours ago
The city by the bay

Or, does denoisePCA() also include runPCA() ?

Yes, it calls runPCA,ANY-method from BiocSingular (not runPCA,SingleCellExperiment-method from scater). You should be able to see a new "PCA" field appear in the SCE returned by this function.

shall the output of plotReducedDim() and plotPCA() look the same ?

Depends on which reduced dimensions you choose to plot in plotReducedDim(). If you set use_dimred="PCA" and you already have PCA results computed in your SCE, then yes, they will be the same; aside from the percentage of variance explained, which is automatically shown by plotPCA() but will needs to be manually specified in percentVar= for plotReducedDim(). If you don't have PCA results already computed in the SCE, plotPCA() will compute them for you but plotReducedDim() will just fail.

does plotTSNE() need runTSNE() in a previous step ?

No, if you plotTSNE() without any "TSNE" coordinates in your SCE, then it will automatically compute the t-SNE coordinates and plot them. So it doesn't need runTSNE() to be run first; but it certainly helps if you're going to replot the same thing over and over (e.g., with different colours). In that case, you can runTSNE() once and then just plotTSNE() with different colouring parameters for rapid plotting.

0
Entering edit mode

thanks a lot, Aaron ! it helps tremendously ! have a good weekend !

0
Entering edit mode

Dear Aaron, if I may add and ask please :

is there is any way to overlay the cluster number on a TSNE plot ?

(or to use a specific list of colors for the list of clusters shown on the TSNE plot). thank you !

0
Entering edit mode

See the text_by= argument in plotReducedDim().

0
Entering edit mode

thank you ! yes, it would be really great to have this info in the package documentation ... thanks ;)

0
Entering edit mode

Well, it is documented, ?plotReducedDim.

0
Entering edit mode

yes, in my R/BioC installation, i was able to see in the ?plotReducedDim documentation :

"colour by","shape by", "size by" (and that is great), not a "text by" entry yet ;)