How do I rescale Golub data before clustering genes?

Question

Suppose I want to cluster genes from the Golub dataset according to their expression profile. Note that I want to specifically cluster genes, and not patients.

It's advisable to do scaling before clustering if distance metric is the euclidian distance. In this case I'd do something like:

data(golub, package = "multtest")
golub_df <- data.frame(golub)
colnames(golub_df) <- factor(golub.cl, levels = c(0, 1), labels = c("ALL","AML"))

golub_scaled <- scale(golub_df)
colMeans(golub_scaled)

The last command outputs values near 0 for each column, which are patients. Is this correct, or should the mean for each gene across the patients be 0(i.e. rowMeans(golub_rescaled) give zeros)?

Thanks in advance!

Answer 1

Scaling is done on the column. It subtracts the mean and divides by the standard deviation , so you should get colMeans(golub) to be around zero.

However you don't need to scale, if you check the vignette:

library(multtest)
?golub

Gene expression data (3051 genes and 38 tumor mRNA samples) from
 the leukemia microarray study of Golub et al. (1999).
 Pre-processing was done as described in Dudoit et al. (2002). The
 R code for pre-processing is available in the file <URL:
 ../doc/golub.R>.

So looking at golub.R, lines 39-41, there is a scaling done:

# Normalization 
golub.expr<-scale(golub,T,T)
dimnames(golub.expr)<-list(NULL,NULL)

the two T indicate TRUE for centering and scaling. So you don't need to scale your data, it's already done. We read in the data:

data(golub)
colMeans(golub)
 [1] -1.245493e-07  2.851524e-07 -3.474271e-07 -1.409374e-07 -4.654212e-07
 [6]  1.002950e-06 -2.917076e-07 -5.145854e-07 -7.604064e-07  4.195346e-07
[11] -5.735824e-07  1.016060e-07  1.507702e-07  6.588004e-07 -1.540479e-07
[16]  5.834153e-07 -1.179941e-07 -6.161914e-07  1.802688e-07  5.735824e-07
[21] -1.606031e-07  1.081613e-07  3.277614e-08 -1.704359e-07 -8.128482e-07
[26] -1.769912e-07 -3.605375e-07 -4.392003e-07  1.179941e-07  5.571944e-08
[31]  6.882989e-08  8.194035e-07  7.014094e-07  6.424123e-07 -7.931826e-07
[36] -4.588659e-08  1.868240e-07  1.179941e-07

Even if you scale it again, it will be unchanged.