Weird confidence intervals from Cuffdiff

[seeker]

Hi All,

Something odd seems to be going on with an analysis I'm doing. I have run the tuxedo suite through to using cummerbund but am seeing some genes which have confidence intervals that do not encompass the FPKM value. The whole confidence interval is lower than the actual FPKM (see attached barplot last pair, and fourth from last pair of bars)

Any ideas what's going on here?

command line call and version info:

 cuffdiff -o diff_out_NA_AA -b Bter_gDNA.fa -p 16 -L NA,AA -u merged_asm/merged.gtf ./tophat_out_CNA/accepted_hits.bam,./tophat_out_DNA/accepted_hits.bam,./tophat_out_ENA/accepted_hits.bam ./tophat_out_CAA/accepted_hits.bam,./tophat_out_DAA/accepted_hits.bam,./tophat_out_EAA/accepted_hits.bam 
version	2.1.1
SVN_revision	4046M
boost_version	104900

[Zapages]

Apparently if you are using Cufflinks 2.1.1 the confidence intervals will be off due to an internal bug. But the actual FPKM values will still be correct.

Hence, I am still using Cufflinks 2.0.2, which from my knowledge does not have this problem.

I hope this helps.

[seeker]

Wow, really? Ugh.

Thanks for that. Can you point me to where you found that info or if you found some other work around? I guess it's not the end of the world to rerun it but it is annoying.

[Zapages]

I read it on seqanswers: http://seqanswers.com/forums/showthread.php?t=39758

I was experiencing the same issues as you were by looking through the expression results from Cufflinks and was very concerned as to why confidence values were off.

Also when you continue on to Cuffmerge and Cuffdiff. Please use the same version ie.if you used 2.0.2 for Cufflinks then use 2.0.2 version for Cuffmerge and Cuffdiff. Likewise for Cufflink/Cuffmerge/Cuffdiff 2.1.1.

All the best with your project.

[seeker]

I searched and didn't see that. Thanks for pointing me to it. Fingers crossed that the next version arrives fairly quickly.

[seeker]

At the risk of hijacking my own thread. Does anyone know what's going on when the expression values are associated with more than one locus? eg. LOC100648144,LOC100648258 in the barplot? They are genes that are right next to one another.