Obtaining DESeq2 LFC values for Loadings of Principal Components
0
0
Entering edit mode
16 hours ago
Nishant • 0

Hi all,

Apologies in advance if this has already been asked previously but I couldn't find a suitable answer.

Context: I am in the middle of analysing my RNAseq data for 80 samples (16 treatments x 5 biol. replicates each) from a marine alga. My treatments were: Temperature as a factor (3 levels), changing carbonate chemistry (manipulated via changing pH and total dissolved inorganic carbon concentration). Although many of carbonate chemistry parameters are highly colinear, I would like to obtain Log2FoldChanges in gene expression due to, for example, CO2 concentration.

Therefore, I thought that the best DESeq2 design model would be to do a PCA on my carbonate chemistry variables, (where PC1 + PC2 explained ~ 95% of variance). So, I used ~ Temp + PC1+ PC2 + Temp:PC1 + Temp:PC2 as my model design, and used the loadings to obtain LFCs for each of my carb. chem. variable.

L1 <- loadings_matrix$PC1[loadings_matrix$variable == var]
L2 <- loadings_matrix$PC2[loadings_matrix$variable == var]
composite_LFC = log2FC_PC1 * L1 + log2FC_PC2 * L2,
composite_LFC_SE = sqrt((lfcSE_PC1^2 * L1^2) + (lfcSE_PC2^2 * L2^2)),
padj_zscore = p.adjust(2 * pnorm(-abs(composite_LFC / composite_LFC_SE)), "BH")

where var is my centered and scaled carbonate chem. variable.

I am not a statistician, or a "hard core" Biologist, but by looking at the code snippet above, do you guys think I'm doing the "correct thing"? I pre-filtered my genes by FDR < 0.05 before running the above code, and I only use the genes which are present in both the wald-test FDR < 0.05 as well as LRT FDR < 0.05 (reduced model for PC1 = ~ Temp + PC2 + Temp:PC2).

Any help/advice/recommendations would be highly welcome and I am grateful for your help.

Nishant

RNAseq Model DESeq2 Design PCA • 104 views
ADD COMMENT
0
Entering edit mode

Therefore, I thought that the best DESeq2 design model would be to do a PCA on my carbonate chemistry variables, (where PC1 + PC2 explained ~ 95% of variance). So, I used ~ Temp + PC1+ PC2 + Temp:PC1 + Temp:PC2 as my model design, who gave you this concept?

and what kind of feature values you are using? for deseq2? raw counts or normalized?

ADD REPLY

Login before adding your answer.

Traffic: 7294 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6