Question: (Closed) Translational Efficiency in 2x2 Experiments Using DESeq2
gravatar for kyusikkim
10 months ago by
kyusikkim0 wrote:

Hello All!
As statistics is not my strong suit, I had some questions involving analyses of our ribosomal profiling and RNA-seq data in a 2x2 experiment. (two strains, treated v untreated). 

  1. In brief, our experiment is looking at the translational efficiency (TE) of the transcriptome when translation is disrupted. As such, we expect to see a global decrease in counts of most genes in our ribosomal protected (RP) reads. What kind of normalization would one use for these conditions? (perhaps checking the distribution of counts to ensure it follows a negative binomial?)
  2. In addition to normalization, would filtering be a potential option? Is there some sort of "good" baseCounts filter ? My guess is that more specific normalization might help this problem.
  3. Reading other posts, the test of the ratio of ratios is the interaction term for a two factor experiment. How would one apply this for a 2x2 experiment? (e.g. fold change in the TE of the interaction effect of treatment and strain). Is this just performing the same analyses for the interaction effect (~ strain + treatment + strain:treatment) except passing a matrix of FPKM-RP/FPKM-mRNA values to DESeq2? 

Any insights or help would be much appreciated!

ADD COMMENTlink modified 10 months ago by Michael Love19k • written 10 months ago by kyusikkim0

Hello kyusikkim!

We believe that this post does not fit the main topic of this site.


For this reason we have closed your question. This allows us to keep the site focused on the topics that the community can help with.

If you disagree please tell us why in a reply below, we'll be happy to talk about it.


ADD REPLYlink written 10 months ago by kyusikkim0
gravatar for Michael Love
10 months ago by
Michael Love19k
United States
Michael Love19k wrote:

Have you seen this post for looking at testing enrichment with DESeq2:

DESeq2 testing ratio of ratios (RIP-Seq, CLIP-Seq, ribosomal profiling)

This approach has been used by many groups, with DESeq2 or edgeR, or similar NB models with an interaction term.

For (2), the filtering in DESeq2 is performed automatically to maximize sensitivity, see the vignette.

For (3), you should just provide all the counts, and let the interaction design take care of the ratios. It's not possible to provide ratios to DESeq2, it will insist on counts.

ADD COMMENTlink written 10 months ago by Michael Love19k


Thank you for the quick response!

I don't think I clearly phrased my question. Below is the experimental design

Treated S1.T S2.T
Untreated S1.U S2.U
Treatment/Strain Strain 1 Strain 2

We are interested in comparing S2.T to S1.U while controlling for the effects of treatment and strain. If my understanding is correct, for comparing counts we just need the interaction effect: ~Strain + Treatment + Strain:Treatment. If we were comparing the same groups, except trying to test differences in TE, how do I model that additional interaction term? Does this mean I test the TE for each assay individually and then compare the two (calculate Strain:Treatment for Ribosome Profiling and Strain:Treatment for RNA-Seq and then compare), change my design to a 2x2x2 experiment (Strain x Treatment x Assay), or somehow group my factors to keep things 2x2 (e.g. S1.U, S1.T, S2.U, S2.T x RNA-Seq, Ribosome Profiling)?

Hopefully that clears things up. If the answer is too complicated for the forums, I will definitely try to delve deeper into the background stats. 


ADD REPLYlink modified 10 months ago • written 10 months ago by kyusikkim0

Can I try to rephrase? You want to know if the TE difference due to treatment is different across strain? If so that would be an interaction model between assay, strain and treatment, yes. I prefer to write out all the coefficients, so it would be: ~strain + treatment + assay + strain:treatment + strain:assay + treatment:assay + strain:treatment:assay. Then you would test on the interaction term. I'd recommend a LRT with a reduced model not containing the last interaction term.

ADD REPLYlink written 10 months ago by Michael Love19k

Sorry for the late response, 

Thank you so much for the help!

ADD REPLYlink written 10 months ago by kyusikkim0
Please log in to add an answer.
The thread is closed. No new answers may be added.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 184 users visited in the last hour