Welcome to the MapMan Family of Software Forum

Please do not hesitate to register and post your question.

Don't forget to subscribe to your posted message so you get notified on updates.
Every question you post will help others and or enhance the software!

Post a question,   post a bug!

Welcome to the MapMen Family of Software Forum Welcome to the MapMen Family of Software Forum

Using MapMan

Small DEGs in RobiNA

Small DEGs in RobiNA
Answer
5/28/14 10:41 PM
Hi Mapmen,

I have been using RobiNA for my RNA-Seq data set. By using the DESeq for the duplicate samples between treatment group and control group, I have got so small number of DEGs (< 10). Here is my detail condition for the run.
P-value cutoff:0.05, P-value correction: BH, log-fold chang min=1: activated, dispersion=pooled.
It looked too stringent. How could I get more DEGs? FYI, by RPKM method with the condition (p-value :0.05, FDR:0.001), more than 300 of DEGs were obtained.

Thank you in advance for your reply.

Jin

RE: Small DEGs in RobiNA
Answer
5/28/14 10:56 PM as a reply to jinhyoung Kim.
Hi

unfortunately I don't know what the RPKM method is.However I sounds as the data would be adultarated already.

Currently most reliable methods -in my opinion- use the actual counts in some way (or keep track of them), as your variance is dependent on count numbers. We reckon the bests ways to analyze data is edgeR and DEseq. You might want to change some of the filters [e.g. remove log fold change=1] or take the actual R scaffold code that robina produces.
(I haven't looked into Gordon Smyths voom transformation but expect that this will perform well, too. Also Bayseq might be ok)

By the way the number of called DE genes is not an indication of the quality of a method, unless you know that these genes should really be differentially expressed.

Best
Björn

RE: Small DEGs in RobiNA
Answer
5/28/14 11:28 PM as a reply to Björn Usadel.
Hi Björn,

Thank you for your reply. RPKM is actually not a specific method, but the way of normalization by read number and gene length. I have already tried with non acivation of log-fold chang min=1, and got a same result. I am also considering to use R code to edit the P-value cutoff. Alternatively, I have performed edgeR with the raw count table obtained from the DESeq. Absolutely, larger number of DEGs as almost ten times than DESeq arose. I am not good at the difference between DESeq and edgeR, I am afraid that the result were too much different. Do you have any experience like this? Thanks.

Jin

RE: Small DEGs in RobiNA
Answer
5/29/14 11:26 AM as a reply to jinhyoung Kim.
Hi Jin

ok simple RPKM for expression estimates, I knew of course, but then you really shouldn't use this (unless you back-calculate) for the reasons I explained.
Between edgeR DeSeq the actual stats are very similar (neative bionomial) but the prior but the normalization has some difference. It sounds like an interesting dataset.
You might want to plot of log fold changes between the two methods to be able to ensure that nothing cheeky happens at the beginning influence the change estimates.
(Btw are you using a simple design A vs B or is this more complex)

Cheers
Björn

RE: Small DEGs in RobiNA
Answer
5/30/14 3:13 AM as a reply to Björn Usadel.
Hi Björn,

Thanks again for your reply. Now I am trying to understand the differenct between edgeR and DESeq. I attached both MA plot from each method. I am still supprised that there showed the great number of differences even they uses same stat. FYI, also I attached raw count table for this comparison. Re my experimental design, there are four different geno type of fish (NT, DF, TR, TF), and one (NT) of them is control.

Cheers,

Jin