Should I do normalization to gene/features or samples?

Question

[SupBioInformatics]

Hi,

I am learning pre-processing metabolomic LC/MS data. For example, I have a data, df, and it has features in rows, and samples in columns. If I want to do z-score normalization, should I do to each sample group or feature?

I am thinking on each features, but I am not sure. For the code in R, I use scale(),

apply(df, 1, function(x) scale(x, center = TRUE, scale= TRUE))

Should I do similar to RNAseq data as well, either to gene or sample?

Thank you.

Answer 1

[khorms]

Normalize columns (samples), not rows (features), because what you are aiming to find is the features with large changes between the sample groups, and you will kill the magnitude of changes if you normalize by features

[SupBioInformatics]

Thanks for the answer.