Question

RNA-seq analysis across different experiments

1

Entering edit mode

mrigaya.mehra ▴ 10

@mrigayamehra-12427

Last seen 7.4 years ago

Hi all,

This is a very general question about an analysis I have been trying to do for quite sometime, but I still do not know how to proceed. I have a series of experiments performed on a single organism which include experiments like: comparing antibiotic sensitive and antibiotic resistant strains, comparing different motility adaptations, comparing stress related adaptations. However, I want to find a set of genes that are commonly dysregulated in all these experiments. I am not sure that how would I find commonly dysregulated genes in these different experiments. I could either just compare the DE genes obtained in each expe and find common or I could use the count files of these experiments together in a single DDS object. And even if I do use their counts together in a single DDS object, how would I normalize them.

Any help would be highly appreciated.

Thanks

limma deseq2 normalization degenes design and contrast matrix • 3.1k views

ADD COMMENT • link written 7.3 years ago by mrigaya.mehra ▴ 10

0

Entering edit mode

Hi Michael,

Sorry for a confusing question. I need to identify genes which show differences in all groups. Like a set of common DE genes between

a) antibiotic sensitive and antibiotic resistant strains DE(a)

b) different motility adaptations DE(b)

c) stress related adaptations DE(c)

So I want something like this DE (abc)

Thanks

ADD REPLY • link 7.3 years ago mrigaya.mehra ▴ 10

0

Entering edit mode

We don't have anything like this in DESeq2, where you have an omnibus test for requiring there to be significant differences in three pairs. What I recommend is to simply compute the intersection of the 3 FDR bounded sets and report clearly to collaborators or in a paper that you looked at the intersection of 3 FDR bounded sets. The final set doesn't have an FDR bound though. You may get different advice from other package authors though.

ADD REPLY • link 7.3 years ago Michael Love 43k

score 1 · Answer 1 · 2017-08-02

It is usually best to keep the experiments separate, because of the difficulties you are already aware of (normalization etc). What people generally do is to simply get DE lists from each experiment and overlap them. Michael Love has already told you the same thing.

A step up from that is to use barcode plots and gene set tests to test whether the different experiments give correlated patterns. We use this technique a lot in our published papers. These methods take a gene list from one experiment, with fold changes attached perhaps, and test for the same expression signature in the new experiment. This technique can be used to correlate any two technologies, the gene list could be from ChIP-seq for example. See ?barcodeplot.

score 0 · Answer 2 · 2017-08-02

hi,

You've included limma and DESeq2 tags, so you're pulling in multiple package authors, so I'll give a software-generic response. You need to provide more specific information about which genes you are seeking. You have different groups of samples. Do you want to find genes which show a difference in any group? To perform testing, you have to be quite specific about what kind of differences you are looking for.

score 0 · Answer 3 · 2017-08-08

There is a method named CORal to compare ranked gene list that would be useful to compare results of different experiments once fold changes were shrunken.

https://www.ncbi.nlm.nih.gov/pubmed/23675929

Other downstream list comparison tools:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2896202/

https://academic.oup.com/nar/article-pdf/42/W1/W161/7437343/gku331.pdf