[BioC] Weights in limma package (theoretical question)

[Gordon K Smyth]

> Date: Tue, 19 Jul 2005 10:37:42 +0200
> From: Matja? Hren <Matjaz.Hren at nib.si>
> Subject: [BioC] Weights in limma package (theoretical question)
> To: <bioconductor at stat.math.ethz.ch>
>
> To all users of limma,
>
> We are using limma package for differential gene expression analysis. We also use weights (0 and
> 1) based on quality control from image analysis. We know that spots with weight 0 are not included
> into normalisation of data (although the MA values are calculated for them also - as explained in
> Limma User's Guide). Our question is what happens with these "0-weight" spots during further DE
> analysis in limma package?

By default, such spots will be ignored in the DE analysis conducted by lmFit() etc.

> Is there any chance that they might appear in the toptable? What to do in that case (simply ignore
> them)? Or do you suggest filtering them out after normalisation?

Your question is not well-posed, because weights apply to *spots*, while it is *probes* which
appear in topTable.  There are a whole string of spots belonging to each probe, one or more for
each array in the experiment, so a spot does not equal a probe.

A probe can still show up near the top in topTable if some of its spots are given zero weight.  On
the other hand, if all the spots corresponding to a given probe are given weight zero, then that
probe will get a NA p-value and will definitely not show up in topTable.  This is the intended
behaviour.

Gordon

> Matjaz and Spela
>
> National Institute of Biology, Ljubljana, Slovenia