Dear Bioconductor community,

after normalizing & filtering an microarray expression dataset, with 11658 rows(probesets) and 26 columns(samples) remained, i tried to do a PCA analysis, Firtsly i transposed the data: data <- t(exprs(eset)). 

But when i used the function prcomp(data, scale=TRUE) i get the following error:

"Error in prcomp.default(data, scale = TRUE) : cannot rescale a constant/zero column to unit variance"

Afterwards when i didnt use the scale argument the pca function worked, but when i used the ggbiplot function, i also get another error:

"Error: invalid rot value"

Is this mistake also related to the argument scale? how can i fixed it ? I also searched some other similar threads for removing columns with zero variance, but it didnt worked. Any ideas or suggestions ??

Please note that neither prcomp() nor ggbiplot() are part of Bioconductor, so this is really not the forum for your question.

That said, is there something about the error message that isn't clear? It says that you cannot rescale a column with zero variance to unit variance (in other words, to convert to unit variance, you divide the values in the column by the standard deviation of the column. If the column has variance zero, the standard deviation is zero, and we all know what happens if you divide by zero). Since you are transposing, this means you have genes (rows) that don't change expression.

It's simple to fix this:

library(genefilter)

ind <- rowVars(exprs(eset)) < .Machine$double.eps

eset <- eset[!ind,]

z <- prcomp(t(exprs(eset)), scale = TRUE)