Diffbind low p-value but high FDR

Question

[LC]

I guess my issue is related to this post https://support.bioconductor.org/p/85487/#85490.

Here is the

dba.report(DBA,th=1, bCounts=TRUE)

results.

One of the peaks clearly shows a significant difference in IGV (And we also expected it to be changed) and has a small p.vaule but the FDR is 1.

seqnames start end width strand Conc Conc_Treatment Conc_Control Fold p.value FDR Control_Treatment_1 Control_Treatment_2 Control_Treatment_3 Control_1 Control_2 Control_3

chr15 99578776 99578976 201 * 5.567216 6.19943 4.415398 1.784032741 3.27E-11 1 74.65 86.82 58.99 27.56 19.1 17.36

Three replicates in the treatment group have more reads (normalized by Deseq2 using RiP) than all three replicates in the control group. p.value is very small but FDR is 1........Would love to see if there is a way to address this issue. Rory Stark

Answer 1

[Rory Stark]

I believe the answer to your question is in that Bioconductor post (https://support.bioconductor.org/p/85487/#85490). DESeq2 is setting the adjusted p-value to NA based on their independent filter, which DiffBind reports as 1.0 for compatibility reasons. You could bypass what DESeq2 is doing (I don't think they are using genefilter::filtered_p() any more) and apply your own multiple testing correction directly to the distribution of p-values, ignoring the independent filtering on mean values.

[LC]

Thank you for your prompt response! It worked out using p.ajust.