Question: Pairwise comparisons in Differential Expression
0
2.2 years ago by
Ron990
United States
Ron990 wrote:

Hi all,

I have 4 RNAseq samples, both at Day 0 and Day 3. So total 8 samples.

I want to do differential Expression analysis between Day 0 and Day 3,but pairwise.

I wanted to confirm if this is The best way to do this is using multi-factor design ? from DEseq2 package. The condition of interest should go at the end of the design formula, e.g. ~ subject + condition. Does my metadata look fine for multi-factor design ?

``````Sample  Subject Condition
Sample1_day0    Day0    Rep1
Sample2_day0    Day0    Rep2
Sample3_day0    Day0    Rep3
Sample4_day0    Day0    Rep4
Sample1_day3    Day3    Rep1
Sample2_day3    Day3    Rep2
Sample3_day3    Day3    Rep3
Sample4_day3    Day3    Rep4
``````

Also, When we do differential expression of 1 group vs the other normally (without multicondition), we get results in the form of 1 log foldchange column ,1 p value column etc. How would we get the results in this ?

Thanks,

Ron

rna-seq deseq • 1.4k views
written 2.2 years ago by Ron990

Your metadata does not look like a multi-factor design. It looks like an experiment with a single factor (Day) with two levels (Day 0 and Day 3). Does "Rep" stand for replicate, or is it a designation of paired samples between the days? The model you mention (~ subject + condition) would I believe not be fittable given you have one observation per subject:condition combination. I guess what you want is a test between days, paired by your "Condition" column. If using a t.test, that could be accomplished in R by (for example) t.test(..., paired=TRUE).

Rep1 is an annotation I added to make sure that Sample1_day0 and Sample1_day3 are replicates since they need to be paired, while doing pairwise comparison.

OK, in that case the model (~subject + condition) will fit a two factor model for each gene. Interpretation would depend on if your condition column is technical factor you are looking to normalize away, or if it is a biological factor of interest that you are trying to quantify. What is your main interest, the differences between Day0 and Day3, or the differences among Rep1, Rep2, Rep3, and Rep4? You would extract the corresponding effects (logFC) and p-values from the two factor model. A good idea is to consult a statistician if you are not clear on the best way to model the data.

I was looking at this post's answer and think this is what I want too:

Deseq2 pairwise comparision