DEG, wilcoxon, p value adjustment

Question

[english.server]

Hi!

A multipart question.

I have exracted log2 transformed normalized TCGA expression data for ~10 genes of my interest. How can I determine wether the genes are differentially regulated between the subtypes of cancer?

1) I think Anova/Kruskal wallis + post hoc for each gene followed by FDR adjustment be ok? As bayesian methods (limma) which borrow info across genes might not well suited for this small number of genes?

2) Furthermore, in genetics filed, and in studies with 2-4 genes, I have hardly seen p value adjustments, which affects decision about borderline p values. Shall I stick to this practice (I mean is there a scientific reason behind it?)

3) not searched for this on biostars so you may skip this: Yet another question, how is the Number of tests for anova+ post hoc determined to determine adjusted p vals?

Answer 1

[German.M.Demidov]

I would go with Kruskal-Wallis + post-hoc. You are right, in my opinion using Limma may be an overkill.

I have hardly seen p value adjustments, which affects at least borderline p values.

You can skip the correction, but you need to understand the consequences: your FDR will be different from 0.05.

how is the Number of tests for anova+ post hoc determined to determine adjusted p vals?

For each ANOVA run you compare all the groups vs all others, so the number is roughly (number of groups) * (number of groups) / 2 - number of groups // since you don't compare a group against itself and when you compare A vs B you don't need to compare B vs A

[english.server]

Dear @German.M.Demidov, Thank you very much for your response.