HI!
I am using DEXSeq to compare exons within two conditions. The applied strategy was the same as in tutorial.
However, after running testForDEU(...) function the following messages were reported:
using supplied model matrix
5 rows did not converge in beta, labelled in mcols(object)$fullBetaConv. Use larger maxit argument with nbinomLRT
I tried to set maxit value, however this did not work:
Error in testForDEU(dxdFull, maxit = 500) : unused argument (maxit = 500)
Afterwards I tried to run next step and got many warning messages:
Fit for gene/exon ENSG00000241769.3+ENSG00000241489.3+ENSG00000010404.13+ENSG00000238039.1 threw the next warning(s): the matrix is either rank-deficient or indefinite
Why is this occurring? How is it possible to fix?
Also, I did not clean up the exon counts before running DEXSeq. Should I perform some filtering in advance (i.e. exclude lowly expressed genes)?
Hi Konstantin. The first one should be a warning and not an error, right? For those 5 exons you will get NAs instead of p-values. It is a problem of convergence of the GLMs. It is likely that these exons have very low counts or very dispersed counts. As for the estimateExonFoldChanges, its the same problem, do you get many NAs in the fold changes column?
Hi Alejandro, number of NAs was 68028 (~11%) for p-value and 70609 (~12%) for log2fold_conditon (total number of exons 588569). Indeed, I did not perform any initial filtering like excluding not-expressed genes. Is there a suitable strategy for the filtering before analysis?
Hi,
I'm trying to use DEXseq too, but I've got the same problem :// Did you solved it?
Thanks a lot,
Nicolas