Hey,

I am currently analyzing some data with DSEq and I want to use the lfcShrink function on 2 different comparisons. However, one of them can only be retrieved from the results using contrast. I am now hesitant to use lfcShrink on this comparison, as I cannot use the coef() argument. Is it valid, to just use the contrast argument as I did it in the results? Here is the code

design2 <- as.formula(~cell_group)
modelMatrix2 <- model.matrix(design2, data = metadata6)
modelMatrix2
Show in New Window
               (Intercept) cell_groupDC2_TLR cell_groupDC3_ctrl cell_groupDC3_TLR
DC2-CD5pos-d1            1                 0                  0                 0
DC2-CD5pos-d2            1                 0                  0                 0
DC2-CD5pos-d3            1                 0                  0                 0
DC2-CD5pos-d4            1                 0                  0                 0
DC3-d1                   1                 0                  1                 0
DC3-d2                   1                 0                  1                 0
DC3-d3                   1                 0                  1                 0
DC3-d4                   1                 0                  1                 0
DC2-BTLA-S-d1            1                 1                  0                 0
DC2-BTLA-S-d3            1                 1                  0                 0
DC2-BTLA-S-d4            1                 1                  0                 0
DC3-CD163-S-d1           1                 0                  0                 1
DC3-CD163-S-d3           1                 0                  0                 1
DC3-CD163-S-d4           1                 0                  0                 1
attr(,"assign")
[1] 0 1 1 1
attr(,"contrasts")
attr(,"contrasts")$cell_group
[1] "contr.treatment"


ddsObj.raw2 <- DESeqDataSetFromMatrix(countData = countdata,
                                     colData = metadata6,
                                     design = design2)
ddsObj2 <- DESeq(ddsObj.raw2)
resultsNames(ddsObj2)
[1] "Intercept"                       "cell_group_DC2_TLR_vs_DC2_ctrl"  "cell_group_DC3_ctrl_vs_DC2_ctrl" "cell_group_DC3_TLR_vs_DC2_ctrl" 

#These are my two desired comparisons
rDC3vDC2_c<-results(ddsObj2,
                    name="cell_group_DC3_ctrl_vs_DC2_ctrl",
                    alpha=0.05)
rDC3vDC2_t<-results(ddsObj2,
                    contrast=c("cell_group","DC3_TLR","DC2_TLR"),
                    alpha=0.05)

So for rDC3vDC2_c i can easily use lfcShrink, as it is one of the resultsNames. However, rDC3vDC2_t cannot be used from ne resultsNames. Are the following lines of Code correct? I am getting an error

ddsShrink_c <- lfcShrink(ddsObj2, coef="cell_group_DC3_ctrl_vs_DC2_ctrl")
ddsShrink_t <- lfcShrink(ddsObj2, contras=c("cell_group","DC3_TLR","DC2_TLR"))

Fehler in lfcShrink(ddsObj2, contrast = c("cell_group", "DC3_TLR", "DC2_TLR")) : 
type='apeglm' shrinkage only for use with 'coef'

People have discussed this issue here: DESeq2 lfcShrink() usage of coef vs. usage of contrast I am however not sure, what to do. I am pretty new to bioinformatics (I am a physician), so hopefully someone can give me an advice. Cheers!

EDIT: I also tried to relevel, however it did not change my resultsNames output:

relevel(ddsObj2$cell_group, ref="DC2_TLR")
ddsObj3 <-DESeq(ddsObj2)
resultsNames(ddsObj3)

 [1] DC2_ctrl DC2_ctrl DC2_ctrl DC2_ctrl DC3_ctrl DC3_ctrl DC3_ctrl DC3_ctrl DC2_TLR  DC2_TLR  DC2_TLR  DC3_TLR  DC3_TLR  DC3_TLR 
Levels: DC2_TLR DC2_ctrl DC3_ctrl DC3_TLR
using pre-existing size factors
estimating dispersions
found already estimated dispersions, replacing these
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing
[1] "Intercept"                       "cell_group_DC2_TLR_vs_DC2_ctrl"  "cell_group_DC3_ctrl_vs_DC2_ctrl" "cell_group_DC3_TLR_vs_DC2_ctrl"

Edit":

ddsShrink_t <- lfcShrink(ddsObj2, contras=c("cell_group","DC3_TLR","DC2_TLR"), type="ashr")

It now worked with ashr. But is this correct?

You can

1) just use 'ashr' for shrinkage, it's also a good method 2) use the technique in the vignette in the shrinkage section, which directly addresses this case. it's one step and provides other contrasts as coefficients