NaN result with DEXSeq

[thurisaz]

I'm having some trouble getting running DEXSeq. I generated count files using the python scripts and have gotten as far as trying to fit the dispersion function, which fails with an error:

> a186Exons
ExonCountSet (storageMode: environment)
assayData: 170526 features, 4 samples
element names: counts
protocolData: none
phenoData
sampleNames: wt-1 wt-2 a186-3 a186-4
varLabels: sizeFactor condition ... countfiles (5 total)
varMetadata: labelDescription
featureData
featureNames: AT1G01010:001 AT1G01010:002 ... ATMG01410:001 (170526 total)
fvarLabels: geneID exonID ... transcripts (13 total)
fvarMetadata: labelDescription
experimentData: use 'experimentData(object)'
Annotation:

> pData(a186Exons)
sizeFactor condition replicate type countfiles
wt-1 1.9951460 W 1 paired-end wt-1
wt-2 1.2640540 W 2 paired-end wt-2
a186-3 0.8502673 m 1 paired-end a186-3
a186-4 0.5048640 m 2 paired-end a186-4

> a186Exons=fitDispersionFunction(a186Exons)
Error in if (sum(log(coefs/oldcoefs)^2) < 0.005) break :
missing value where TRUE/FALSE needed
In addition: Warning message:
In log(coefs/oldcoefs) : NaNs produced

I'm guessing the error (about missing values) is probably a result of the NaNs produced. estimateDispersions didn't give any warnings or errors. To be honest, I'm not even sure where the NaNs are coming from -- I'm guessing it's not a problem that there are exons with a zero count, right? I'm at a loss about what to try next, so I would be very grateful for a pointer in the right direction (or maybe even a solution). If some extra information would be helpful in sorting this out, please let me know.

Thanks!

[areyes]

fixed!

Hi!

Thank you very much for the error report and from sending me your ecs object. I found the problem and its now fixed in version DEXSeq 0.99.3,
it was a detail when dealing with NAs.

Details, I did not know this happened in R:

> x <- rnorm(100, 0, 1)
> x <- c(x, rep(NA, 2))  ## introduce some NAs
> x[which(x < 0)] == x[x < 0]
 [1]  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE
[13]  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE
[25]  TRUE  TRUE  TRUE  TRUE    NA FALSE FALSE FALSE FALSE FALSE FALSE FALSE
[37]    NA FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
[49] FALSE
Warning message:
In x[which(x < 0)] == x[x < 0] :
  longer object length is not a multiple of shorter object length

If there are NAs in a logical vector, subsetting in the below way does not work, a 'which' needs to be added

x[x < 0]

Thanks again!

[thurisaz]

Thanks very much for the quick fix! The Bioconductor website still lists version 0.99.2 (http://www.bioconductor.org/packages...ml/DEXSeq.html). Is there somewhere else to look for the new version or should I just wait for it to be updated there?

[areyes]

You can get it directly from the bioconductor svn repository a indicated here:

http://wiki.fhcrc.org/bioc/SvnHowTo

And install it manually!

[MerFer]

I've installed the fixed version of DEXSeq. But I get the same errors again.

ecsExons <- fitDispersionFunction(ecsExons)
Error in if (sum(log(coefs/oldcoefs)^2) < 0.005) break :
missing value where TRUE/FALSE needed
In addition: Warning message:
In log(coefs/oldcoefs) : NaNs produced

[thurisaz]

Thanks, areyes. I installed the new version and my problem is solved.

[leda]

This is an old thread and it seems the wiki link is broken now... is there another bioconductor svn how-to somewhere?

thanks,
Leda

You can get it directly from the bioconductor svn repository a indicated here:

http://wiki.fhcrc.org/bioc/SvnHowTo

And install it manually!

[areyes]

Google "bioconductor svn repository":

http://www.bioconductor.org/developers/source-control/

[Simon Anders]

No need to use the SVN repository, anyway. Changes get propagated from the SVN repo to the downloadable packages within a few days, and this fix is several months old.