DE analysis model matrix for paired samples
1
0
Entering edit mode
thkapell ▴ 10
@tkapell-14647
Last seen 12 weeks ago
Helmholtz Center Munich, Germany

Hi,

I am analyzing a NanoString dataset where the metadata look as below:

The factor of interest is the "group" and both groups are found in each mouse ("mouseID") (paired samples). I used DESeq2 for the analysis and the PCA on the normalized data looks like this:

It seems like there is a batch coming from the animals in PC1, although PC2 mildly separates the groups. I have set my design as: ~mouseID+group to account for differences between animals in my downstream DGE analysis. I wanted to ask whether you think I should perform a stronger batch correction (e.g. limma). Also, I read a recent article on NanoString analysis where RUVg() was used for normalization. Below I attach how the samples look before and upon normalization with DESeq2 (counts()). Do you reckon I should use RUVseq offset to normalize the samples?



NanoStringNCTools DESeq2 • 288 views
0
Entering edit mode

PC1 is only 12% of the variance? That's not a lot. It looks to me like none of your experimental conditions had much effect at all.

0
Entering edit mode
@mikelove
Last seen 1 day ago
United States

Agree with swbarnes2. It just seems like the condition effect is very small. Also I'd double check the metadata. Sometimes samples are swapped which can lead to strange looking PCA plots.

0
Entering edit mode

Thank you both for your comments. I totally agree that the group effect is quite small (evident in PC2 and PC3). Would you additionally recommend to use another normalization method (e.g. RUVg) to remove the unwanted variance and (perhaps) make the group effect stronger?

0
Entering edit mode

I don’t think that will help but this is up to you.