Deseq2, lfcThreshold .
1
0
Entering edit mode
Iryna • 0
@iryna-17263
Last seen 14 months ago
Sweden

Hi,

I have pretty simple but probably weird question. I have complex design with interaction terms in my model (see below), I run DESeq2_1.20.0 on rna-seq data. It seems that I can't change my contrasts to end up using only apeglm shrinkage and lfcThreshold, instead I'm stuck with using ashr shrinkage method without being able to incorporate lfcThreshold into calculation of null.

What bothers me is that all published papers are using simple comparisons, at least in plant science, but all, test against non-zero log fold change. If me using ashr shrinkage and definitely avoiding doing any post-hoc on lfc, how can I justify not incorporating lfcThreshold?

previously suggested by Michael L. model matrix for my study.

mm <- model.matrix(~cltv + cltv:time + cond:cltv:time, data=colData(dds)) 

Thanks.

ashr Deseq2 lfcThreshold • 176 views
0
Entering edit mode
@mikelove
Last seen 4 days ago
United States

You can either: use an lfcThreshold with apeglm by re-arranging terms such that your contrast is represented by a single coefficient (some strategies are in the vignette), or if this isn't clear how to do and you can't find someone to guide you through re-arranging the coefficients, you can avoid LFC shrinkage and use results() with lfcThreshold, so providing an FDR controlled list of genes with LFC > some value. If you go with the latter, you can just use a higher threshold on counts (e.g. X or more samples with a count of 10 or more) in order to remove the genes with noisy LFC.