Tukey's HSD in edgeR

Question

[ygowtha]

Hello All

I am using edgeR to perform statistical analysis on my RNA-Seq data set. I have been able to successfully perform one-way anova test on my data but now I want to perform post-hoc analysis on the result obtained. I was thinking on the lines of performing a Tukey's HSD test as a part of post-hoc analysis. Can anyone help me with providing the guideline or code to perform the Tukey's HSD on edgeR? Also, I am open to suggestions regarding other alternate post-hoc analysis.

Thank You

Answer 1

[Devon Ryan]

ANOVA-like

You haven't done used a one-way ANOVA since that has no meaning on count-based data. I've not seen anyone come up with a good way to perform a post-hoc test on count-based data like this (see also the "anova like test in edgeR" thread on the Bioconductor list and Ryan Thompson's reply in particular).

If you really want to do a post-hoc test then just use limma instead.

[ygowtha]

Thank you Devon for your reply. Yes, it is ANOVA-like. Since I am novice to the edgeR platform, I would like to know the real difference between ANOVA and ANOVA-Like method. Why do you say it has no meaning on count-based data. I was of the impression that counts are representative of the gene expression and that the means of counts from each group could be used for statistical analysis?

Also, in your opinion, which would be the best platform to perform ANOVA analysis on?

Thank you very much. I really appreciate your help with this.

Regards

[Devon Ryan]

Well, an ANOVA is a special case of a linear model, which assumes Gaussian variance. Count data follows a negative binomial distribution, where variance increases with mean. edgeR uses a generalized linear model and then empirical Bayes variance shrinkage.

As I suggested previously, just use limma/voom.