[BioC] LIMMA: choice of offset value for background correction method "normexp"

[Gordon Smyth]

Dear Pie,

The offset means:

    R = normexp(Rf - Rb) + offset

where normexp() is the transformation used by the 'normexp' 
background correction method. The purpose of the offset is to reduce 
the variability of the log-ratios for low intensity spots, because 
the log-ratios

   M = log2( R / G)

are damped towards 0 by larger offsets. The optimal choice is the one 
which makes the variability of the log-ratios as constant as possible 
across the range of intensity levels. (This is the same general 
purpose as the vsn package, but by other method.)

You can judge a good value for the offset by inspection of the 
MA-plots. If you really want a quantitative way to judge this, look 
at the component fit$df.prior after you use the eBayes() function in 
limma. The better you stabilise the variances, the larger will be 
df.prior and the greater will be the power to detect DE genes. Hence 
the offset which maximises df.prior is, in sense, optimal.

Best wishes
Gordon

>Date: Fri, 31 Mar 2006 16:46:15 +0100
>From: Pie Muller <pie.muller at liverpool.ac.uk>
>Subject: [BioC] LIMMA: choice of offset value for background
>         correction method "normexp"
>To: bioconductor at stat.math.ethz.ch
>
>Dear Bioconductors
>
>I am using the package limma and understand that the type of background
>correction is important in the further processing of the data. Playing
>around with own data I found that using the "normexp" function with
>different values for the offset changes the final outcome of statistical
>testing. I am, therefore, a bit concerned about which is the best way to
>choose a value for the offset in "normexp". In the examples provided I
>found values of 25 or 50 but it was unclear to me why you would choose
>values in that range. Is there any rational in choosing a specific offset
>value other than just "playing around". In fact, what does the offset value
>mean? Offset from the background? Apologies if I didn't understand a simple
>principle here...
>
>Thanks for any help on this issue!
>
>Pie
>
>_______________________________________________________
>
>Dr Pie Mueller
>
>Vector Research Division
>Liverpool School of Tropical Medicine
>Pembroke Place
>Liverpool
>L3 5QA
>UK
>
>Tel. +44 (0)151 705 3123 (office)
>Tel. +44 (0)151 705 3140 (Gwen Finnegan Dept Secretary)
>Fax. +44 (0)151 705 3369
>
>http://www.liv.ac.uk/lstm/