[BioC] limma: instances of highly variable paired ratios, but very small p-values

Gordon K Smyth smyth at wehi.EDU.AU
Sat Nov 17 07:36:42 CET 2012


Dear Ryan,

The limma algorithm is very well understood by now, and there are many 
bioinformatications at the FHCRC who could probably answer your question.

I find it hard to make a response to your email because I just see a 
jumble of numbers.  I don't have a firm understanding of what your 
experimental design is, what model you've fitted, what the numbers are 
that you've calculated, or what you want me to see.  To comment more, I'd 
need to see your experimental design and the complete pipeline of commands 
used to generate the output given.

Best wishes
Gordon

I don't a firm idea of what your experimental design

> Date: Wed, 14 Nov 2012 17:23:43 -0800
> From: Ryan Basom <rbasom at gmail.com>
> To: bioconductor at r-project.org
> Subject: [BioC] limma: instances of highly variable paired ratios, but
> 	very small p-values
>
> Hi,
>
> I've performed a limma analysis on paired samples that were run on 
> Illumina HT12 arrays, with three replicates in each condition.  I'm a 
> bit troubled by the results though, as there are several probes that 
> have very small adjusted p-values, though when looking at the paired 
> ratio values, they vary quite a bit.  Here are a few examples where the 
> comparison is long-short, and the samples are paired by the letters 
> A,B,C.  After the adj.P.Val column, I've calculated the paired sample 
> ratio values, these three columns are followed by the signal intensities 
> from each sample:
>
>    ProbeID TargetID logFC AveExpr P.Value adj.P.Val long-short.A 
> long-short.B long-short.C long.A long.B long.C short.A short.B short.C 
> 1450390 RPL17 -1.3733092649 10.2105020267 4.35314891863083e-17 
> 4.55305891127872e-14 -1.277287712 -0.6714209686 -2.1712191142 
> 9.5618416199 9.5085763086 9.5011242541 10.8391293319 10.1799972771 
> 11.6723433683 1230376 ALAS2 1.4395987013 10.0069363572 
> 4.9058551374517e-17 4.76463659313792e-14 0.356310701 2.2275874983 
> 1.7348979044 9.1085909827 10.4991863428 12.5724297981 8.7522802817 
> 8.2715988445 10.8375318936 3420451 RSL24D1 -1.2585240828 8.0288125742 
> 6.26229691539969e-15 4.73046950881609e-12 -0.9845918613 -0.6335605827 
> -2.1574198045 7.5348228063 7.4471358372 7.2166929547 8.5194146676 
> 8.0806964199 9.3741127592
>
>
> I'd assumed that limma would have been more sensitive to this and am 
> wondering if anyone could please explain why this may be occurring.
>
> Thanks,
>
> Ryan
>
> __
> Ryan Basom
> Systems Analyst/Programmer
> Genomics Resource
> Fred Hutchinson Cancer Research Center
>

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}



More information about the Bioconductor mailing list