I am analysing an rnaseq with interaction terms that shows extreme values on low count genes. As expected, these genes get shrinked to a FC ≈ 0 with
As I understand, all the genes that are now shrinked to a FC of ≈ 0 are genes that did not have enough count information to reliably predict a FC, and therefore get "nullified", but they are still significance (as
lfcShrink does not recalculate significance).
However, I am now dubitative in if I should include these genes in downstream analyses (such as enrichment, pathway, etc.) as I am not sure of what is the biological meaning of their significance.
So here are specific questions:
Should I filter these genes from enrichment analysis, as we don't have enough infromation about them?
Shouldn't these genes be picked by the independent filtering of DESeq2?
Would in this case benefit from a pre-filter to directly eliminate all these genes from the dataset?
Note: I am running DESeq2 as:
ddsTxi <- DESeqDataSetFromTximport(txi, colData = sampleGrouping, design = ~ genotype + treatment + stress + treatment:stress) dds <- DESeq(ddsTxi, fitType = "local") reslfcSh <- lfcShrink(dds, coef = 6, type = 'apeglm')
Thank you very much for your time in advance.