Question: DESeq2: Appropriate way to deal with knockouts in experiment design (RIPSeq)
0
gravatar for snamjoshi87
3.0 years ago by
snamjoshi8730
snamjoshi8730 wrote:

Hello, I am new to DESeq2/RNAseq and would like a little help. I have performed a RIPSeq experiment with conditions described below (mouse neural tissue harvested after drug injections). I am aware that there are alternatives such as RipSeeker, however I am treating my samples as RNAseq for the time being. I have imported my raw counts into R and analyzed them with DESeq2. However, I am unsure how to take my knockouts into account. In this case, the knockouts essentially represent background bead binding. Since RIPSeq tends to be noisy, I need to "subtract" out any RNAs that might bind at a background level. I have read online that I cannot subtract counts but then I am not quite sure how to compare my DESeq2 results data frames between WT and KO.

<caption>Experimental design</caption>
Category Condition Replicates
Control Solvent 1 2
Control Solvent 2 3
KO Solvent 2 KO 3
Experimental Drug 1 3
KO Drug 1 KO 1
Experimental Drug 1 + Drug 2 3
KO Drug 1 + Drug 2 KO 3
Input Total RNA 3

It is also important to mention that since drug 1 and drug 2 have different solvents, I have controls for each separate solvent. The hope is that the two solvents alone will not show a huge difference in what I immunoprecipitated.

Here is my analysis strategy:

  1. Collapse technical replicates (there are two technicals for each biological replicate)
  2. Slice up DESeq2 object into subsets that include the condition as well as the total RNA
  3. Run results on each of these comparing the condition against the total RNA to find enrichment
  4. Compare each condition and its knockout pair to see if any top differentially expressed genes showed up in both experimental and KO
  5. If so, remove gene from experimental condition list

So my question is whether or not this is valid and if there is a better way to do this kind of comparison to deal with the knockouts given my experimental conditions.

ADD COMMENTlink modified 3.0 years ago by Michael Love24k • written 3.0 years ago by snamjoshi8730

First, just adding a comment: usually when people want to "subtract" out background expression, the best approach is to use a design with interaction terms. Interaction terms help you test for differences of differences, for example: is the ratio of KO/WT different with drug 1 compared to control, or drug 1+2 compared to control, or drug 1+2 compared to drug 1?

This is fairly easy to set up, and does not require subsetting the DESeq2 object. You just run DESeq() once and get multiple results tables.

Does this sound like what you are interested in testing?

Some experiments though which I'm not sure the role they play in the analysis are: solvent 1 control, and total RNA. 

ADD REPLYlink written 3.0 years ago by Michael Love24k

I would be interested in setting up something similar to what you described. Would this be like testing the difference between WT and KO within each condition? I think I will have to try this out and see the data before I'd be sure that I am understanding exactly what I would get back. Essentially, I want the DGE for the experimental sample to take into account the fact that the KO gives you a sense of the amount of noise that may exist in the experimental sample.

For solvent 1 control, I think something that could be done is to compare drug1+2 against solvent 1 and then drug1+2 against solvent 2. I was not able to dissolve the two drugs in the same solvent for the drug injections (each was separately dissolved in a different solvent then combined together in the syringe) so I need a way to measure similarity in the results between the fold change relative to each control to show that one solvent wasn't biasing the results I see. This design may need to be different from the KO one described above.

Is there anything wrong with comparing directly to my total RNA as I described above? This kind of analysis is done with qPCR for example where you see relative enrichment against the entire population. Is there no way to factor this into the analysis of WT and KO?

ADD REPLYlink modified 3.0 years ago • written 3.0 years ago by snamjoshi8730

Okay, I think I figured out how to use interactions as you said. In my condition column I added in either CTRL (solvent 2) or drug1 or drug1+2. Then in a genotype column I listed which were WT or KO. I removed the solvent 1 and totals data for now. After running it, I got the following result tables back:

  • Intercept
  • genotype_WT_vs_KO
  • condition_drug1_vs_CTRL
  • condition_drug1drug2_vs_CTRL
  • genotypeWT.conditiondrug1
  • genotypeWT.conditiondrug1drug2

Then I can use:

results(ddInt, contrast = list(c("condition_drug1drug2_vs_CTRL", "genotype_WT_vs_KO")))
results(ddInt, contrast = list(c("condition_drug1_vs_CTRL", "genotype_WT_vs_KO")))

This should give me the differential expression of either the drug1 or drug1/2 v. control controlling for the interaction of WT vs KO. Is this correct?

I feel the usage of total RNA is still valid but it is really just telling me which genes change in expression relative to the whole background population. It doesn't tell me anything about how it changed relative to a condition.

ADD REPLYlink modified 3.0 years ago • written 3.0 years ago by snamjoshi8730
Answer: DESeq2: Appropriate way to deal with knockouts in experiment design (RIPSeq)
2
gravatar for Michael Love
3.0 years ago by
Michael Love24k
United States
Michael Love24k wrote:

Hi,

Switching to answer now. Using these results names:

  • Intercept
  • genotype_WT_vs_KO
  • condition_drug1_vs_CTRL
  • condition_drug1drug2_vs_CTRL
  • genotypeWT.conditiondrug1
  • genotypeWT.conditiondrug1drug2

To test if the WT vs KO effect is different in drug 1 vs control you would use:

results(dds, name="genotypeWT.conditiondrug1")

Similarly to compare drug1+2 vs control you would pull out name="genotypeWT.conditiondrug1drug2". 

To compare drug1+2 vs drug 1 you would use:

results(dds, contrast=list("genotypeWT.conditiondrug1drug2", "genotypeWT.conditiondrug1"))

I'm still not clear how to make use of the Total RNA or Solvent 1 experiments, because the above comparisons are standalone. Adding Total RNA to the numerator and a denominator of a contrast doesn't help because it will just cancel out. And because you are comparing WT vs KO, and there is no KO sample using Solvent 1, I don't see how you can make use of Solvent 1 in the above comparisons. 

 

ADD COMMENTlink written 3.0 years ago by Michael Love24k

This works really well! Thanks for answering my questions and thank you for developing DESeq2.

ADD REPLYlink written 3.0 years ago by snamjoshi8730
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 333 users visited in the last hour