Question

2 factor analysis using DESeq2 but the result is not what I expected from the design

0

Entering edit mode

JY • 0

@f0bb40a8

Last seen 9 days ago

South Korea

Hi, I'm trying to do two-factor DEG analysis using DESeq2.

colData(dds) 

      condition    batch    infection  metabolite  sizeFactor
       <factor> <factor>     <factor>   <factor>  <numeric>
Mock1      Mock   first  Not-Infected        Non   0.904435
Mock2      Mock   first  Not-Infected        Non   1.060912
Mock3      Mock   second Not-Infected        Non   0.864510
Mock4      Mock   third  Not-Infected        Non   1.064498
L1         L    third  Not-Infected        L   1.137483
...         ...      ...          ...        ...        ...
V4     Virus       third     Infected        Non   1.104777
LV1    L+Virus    first     Infected        L   0.862166
LV2    L+Virus    first     Infected        L   0.897168
LV4    L+Virus    third     Infected        L    1.142766
LV5    L+Virus    third     Infected        L   1.160191

What really matters are infection and metabolite columns and I ran the following code.


dds_fin<-DESeqDataSetFromMatrix(countdata_fin,colData = metadata_fin,design = ~ infection + metabolite + infection:metabolite)

To analyze the main effect of infection, which means the effect of infection in non-treated group, I ran the following code.

res_infection<- results(dds_fin,contrast = c("infection","Infected","Not-Infected"))

The results said there is only one gene (GNAT1) that are differentially expressed with statistical significance and high log2FoldChange value (between mock and V group as far as I understand correctly). My question is, if I see the normalized expression of GNAT1 gene like the image I posted, there is almost no difference between Mock and Virus-infected(V) group. I'm really confused with this result. Normalized expression level of GNAT1

And another question is about contrast. What I really want to know is the effect of infection regardless of what is treated (control or metabolite L). I saw deseq2: coding 2x2 design , where situation seems similar. So if I want to extract the pure infection effect, do I need to do numerical contrast?

I appreciate any guidance or advice. Thanks a lot!

JY

DESeq2 • 148 views

ADD COMMENT • link 14 days ago JY • 0

score 0 · Answer 1 · 2024-10-17

For help with statistical analysis and forming contrasts, I recommend working with a local statistician. I have to reserve my time on the support site for software related questions.

Note that the comparison you extract with this design and code depends on the reference level for metabolite.

You can compare your results to what you get with a design of ~condition.