comparing "ashr" vs "Normal" , how to decide which method is suitable for rnaseq analysis or is there other parameters we need to set ?

Question

[majeedmj.ict]

Dear experts,

We compared the transcriptome of cells derived from patients (not cell lines).

In our experiment we have a condition day1 vs day 0 and I have applied DESEQ2 package to perform differential expression. I have applied both “Normal” and “ashr” method for LFC shrinkage . I was wondering which method suits for my analysis. Below I given for your reference MA plot for both “ashr” and “normal” method as shown in figure 1. And I got volcano plot for each “ashr” method and “normal” method as shown in figure 2.

Thresholds used:

Pvalue = 0.05

LFC = 1.0

Rest is default parameters.

I need your opinion or suggestion for my RNAseq analysis, whether what I did was feasible or can you suggest which parameters I have to adjust to get best results ?

Answer 1

[ATpoint]

Mike Love (the DESeq2 author) recommends both ashr and apeglm for shrinkage of the effect size estimates while actively discourages use of normal. This is extensively discussed in their paper that introduced apeglm, see https://academic.oup.com/bioinformatics/article/35/12/2084/5159452

I personally use ashr simply because it can use contrasts whereas apeglm works via the coefficients which can be a bit tedious (type='apeglm' shrinkage only for use with 'coef'). In any case, do not use normal, please see the paper for details. If you browse for threads at support.bioconductor.com you will find many threads with that advise.

[majeedmj.ict]

Thank you :)

[Harish.Sudarsanam]

maybe this is too old to ask a follow up question, but I read that both type "normal" and "apeglm" use posterior mode and "ashr" uses posterior mean. Isn't using posterior mode a better option than mean??, so apeglm should be a better option than asher. what do you think about it??