proteomics data analysis and non-normality distribution

Question

[gogeni5529]

Hi, I'm new to both proteomics and biostar and would like to start with maybe a simple, but a more general question.

I know that for RNA-Seq, people just assume normality to the data mostly, because to achieve normality with such a small number of samples is always risky and difficult. I was wondering how this is handled in the proteomics field. What kind of tests one might use to analyze the data, when a t-test is not applicable due to the non-normal behavior of the data?

What alternatives do I have there?

I saw that some people are using LIMMA. Can LIMMA handle non-normal distributed data?

thanks

G.

Answer 1

[Papyrus]

When using limma in proteomics/metabolomics, the data are usually log-transformed first, which is usually makes it normal/more homoscedastic (or assumed to be so). (Example paper).

By the way, people do not usually assume normality for RNA-seq data. They usually model raw counts with negative binomials (e.g. DESeq2, edgeR) or do some other transformations (e.g. limma-voom).