[BioC] base of B-value log in Limma - still not sure

[Gordon Smyth]

At 08:00 PM 6/05/2006, bioconductor-request at stat.math.ethz.ch wrote:
>Date: Fri, 5 May 2006 11:42:22 -0400
>From: Richard Friedman <friedman at cancercenter.columbia.edu>
>Subject: Re: [BioC] base of B-value log in Limma - still not sure
>To: Morten Mattingsdal <morten.mattingsdal at student.uib.no>
>Cc: "bioconductor at stat.math.ethz.ch List'"
>         <bioconductor at stat.math.ethz.ch>
>
>Dear Morten,
>
>         Thank you for your answer, but I am still not sure based upon it.
>In Limma log2 is used for absolute intensities, and for  M.
>M=log2(Intenisity2)-log2(Intensity1). In limma, B and M appear in
>the same table and graphs. This is unusual if B and M have different
>bases.
>
>         Also, at least in the usage with which I am familiar, "natural: logs
>are
>base e not base 10" and I have seen "log" without a subscript used for
>both base 10 and base e logs. So even if B is not base 2 I am not
>usr eif it is base 10 or base e.
>
>         Would Gordon care to comment?
>
>Thanks and best wishes,
>Rich

The answer to this question is in Section 10.1 ("Output from eBayes") 
of the Limma User's Guide, where it is explained explicitly how to 
convert from the B-statistics into probabilities.

In R and in mathematics, log() always means base-e unless otherwise 
stated. Other disciplines may be less precise, but all math stat 
journals that I know of follow this convention.

The use of base-2 for M-values is a reference to 2-fold-changes, 
which have a traditional meaning. There is no such interpretation for 
probabilities or odds-ratios, hence no reason to use base-2 for the 
B-statistics. In any case, the B-statistic was defined by Lonnstedt 
and Speed (2002), and it is not appropriate for limma to change the definition.

Best wishes
Gordon