Hi,

  I got the following code, and want to generate the heatmap later. 

y <- calcNormFactors(y, method = "TMM")

v <- voom(y, design, plot=TRUE)

lcpm <- cpm(y, log=TRUE)

From EdgeR, it was recommended that lcpm could be used to generate heatmap with a prior.count. However, from https://f1000research.com/articles/5-1408/v1, the tutorial uses v$E to generate the heatmap.

I compared v$E and lcpm here, there is very little differences here: 

max(v$E[,] - lcpm[,]) 

returns 2.9. 

Any suggestions for the heatmap? Which one should I use? ( I tried both, and I saw little difference for my current dataset)

Thanks & Regards,

Raymond

I have always recommended and used cpm() with log=TRUE and prior.count between 2 and 5, for example:

lcpm <- cpm(y, log=TRUE, prior.count=3)

because it reduces further the variability of low-count genes. I would have preferred that in the F1000Research paper that you cite, but, as you have noticed, it may not make much difference. The different choices correspond to different choices for prior.count, and the heatmap is not usually very sensitive to this choice.

I would make the heatmap the same regardless of whether I was using edgeR or limma for the DE analysis.