[BioC] NaN p-values from edgeR

[Jared Bischof]

Thanks Gordon.  I'm not sure what was causing that output but updating
to the newest version of edgeR fixed the problem.

Thanks,
Jared

On Sat, Apr 28, 2012 at 8:20 PM, Gordon K Smyth <smyth at wehi.edu.au> wrote:
> Dear Jared,
>
> edgeR does not generate NaN p-values as far as I know so, no, we have not
> seen this before.
>
> Your output shows that you are not using the current release version of
> edgeR.  Please try installing the current version.
>
> As Wolfgang said in his reply, we like to see sessionInfo() and more
> information about how output was generated.  Making useful comments on
> snippets of output is hard.  Have a look at the posting guide:
>
> http://www.bioconductor.org/help/mailing-list/posting-guide/
>
> Best wishes
> Gordon
>
>> Date: Fri, 27 Apr 2012 14:00:29 -0500
>> From: Jared Bischof <jared at alum.northwestern.edu>
>> To: <bioconductor at stat.math.ethz.ch>
>> Subject: [BioC] NaN p-values from edgeR
>>
>> Hi all, I'm a computational biologist at Children's Memorial Hospital
>> in Chicago.  I'm using the edgeR library to compare RNA-seq datasets.
>> But, recently I was running a comparison and got an error that led me
>> to the realization that some of the p-values were being returned as
>> "NaN".  I searched the message archives and found messages addressing
>> the Fold Change sometimes showing up as NaN but this is the p-values
>> in my case.  I checked the input and neither of the tag counts is zero
>> so I know that's not part of the issue.  Has anyone seen this before?
>> I'm wondering what this represents in this case.  I thought that
>> perhaps this represented a p-value of essentially zero but looking at
>> the input data that is clearly not the case.  Here's part of the
>> counts table:
>>
>>        Tumor3TagCounts Tumor5TagCounts
>> FR138965          171971           27074
>> FR077061          155101           39388
>> FR273624          153178           38882
>> FR121537          100582           83496
>>
>> And here's a relevant section of the table from the exactTest output:
>>
>>          logConc      logFC      p.value
>> FR138965 -4.621888 -0.5596774 0.000000e+00
>> FR077061 -4.425942  0.1301280 1.534497e-50
>> FR273624 -4.444269  0.1294732 1.923549e-49
>> FR121537 -4.196385  1.8389148          NaN
>>
>> I can't reason that NaN represents a p-value of zero because the
>> transcript FR273624 has an even greater difference and does not return
>> a p-value of zero.
>>
>> Thanks in advance,
>> Jared Bischof
>>
>
> ______________________________________________________________________
> The information in this email is confidential and inte...{{dropped:6}}