[BioC] Unexpectedly high FDR using contrasts in Limma?

[Gordon K Smyth]

On Sat, April 7, 2007 12:43 am, J.delasHeras at ed.ac.uk wrote:
>> Other things like including a dye-effect, background correcting, or
>> otherwise checking the data
>> often help give more powerful results.
>
> I thought the dye effect is incorporated by default in limma, without
> specifying it implicitly.

I'm referring to probe-specific dye-effects.  See page 36 of the User's Guide.  limma
automatically takes into account the arrangement of dye swaps when computing the treatment
effects, but it doesn't automatically include probe-specific dye-effects.  In an experiment such
as yours, including a dye-effect term, which is just an intercept term for your experiment, can
globally reduce the standard errors and hence increase statistical significance overall.

> I've been meaning to try single channel analysis in limma. I wonder if
> that could help me improve matters a little, by allowing me to compare
> MBD1 against DBL directly...

Single channel analysis doesn't turn an indirect comparison into a direct one.  That is a
structural matter.  For your experiment, single channel analysis simply recovers information from
the A-values as well as M-values, usually about 8% extra information.

Best wishes
Gordon