Dear all,
I am having issues with my DEXSeq analysis. The symptom appears with plotDEXSeq but I think this is a deeper issue with the underlying computation. The error message I am getting is with plotDEXSeq is:
Error in plot.window(xlim = c(0, 1), ylim = c(0, max(matr))) :
need finite 'ylim' values
The issue is that in my Exon Count set object the log2fold values are all either NA or NaN:
head(featureData(DexSeqExons.loc)@data$log2fold)
[1] NaN NaN NaN NaN NaN NaN
I suspect, but I am not sure, that the following warning points to the problem:
In fitDispersionFunction(DexSeqExons.loc) :
Negative intercept value in the dispersion function, it will be set to 0. Check fit diagnostics plot section from the vignette.
Indeed I see:
> DexSeqExons.loc@dispFitCoefs
(Intercept) I(1/means[good])
0.0000000 0.9973345
which does not seem right. I can somehow get around the error by changing the function fitDispersionFunction and set
coefs[1] <- 0.001
instead of
coefs[1] <- 0
when the estimated coefficient is negative. But tinkering with the underlying numerical computations hardly seems like a good thing to do.
I had never seen such an issue before, and I had been working on the same dataset. It is not clear to me what I may have changed that caused this odd behaviour but I cannot find a way to fix it. There could be something wrong with my gff file which I changed recently but I don't think it is very likely.
Has anyone experienced a similar issue before? Any tip to fix the problem would be very welcome,
Thanks,
Vincent
I am having issues with my DEXSeq analysis. The symptom appears with plotDEXSeq but I think this is a deeper issue with the underlying computation. The error message I am getting is with plotDEXSeq is:
Error in plot.window(xlim = c(0, 1), ylim = c(0, max(matr))) :
need finite 'ylim' values
The issue is that in my Exon Count set object the log2fold values are all either NA or NaN:
head(featureData(DexSeqExons.loc)@data$log2fold)
[1] NaN NaN NaN NaN NaN NaN
I suspect, but I am not sure, that the following warning points to the problem:
In fitDispersionFunction(DexSeqExons.loc) :
Negative intercept value in the dispersion function, it will be set to 0. Check fit diagnostics plot section from the vignette.
Indeed I see:
> DexSeqExons.loc@dispFitCoefs
(Intercept) I(1/means[good])
0.0000000 0.9973345
which does not seem right. I can somehow get around the error by changing the function fitDispersionFunction and set
coefs[1] <- 0.001
instead of
coefs[1] <- 0
when the estimated coefficient is negative. But tinkering with the underlying numerical computations hardly seems like a good thing to do.
I had never seen such an issue before, and I had been working on the same dataset. It is not clear to me what I may have changed that caused this odd behaviour but I cannot find a way to fix it. There could be something wrong with my gff file which I changed recently but I don't think it is very likely.
Has anyone experienced a similar issue before? Any tip to fix the problem would be very welcome,
Thanks,
Vincent
Comment