[BioC] Limma - using both array and spot weights in lmFit

[Gordon K Smyth]

> Date: Wed, 2 May 2007 15:04:26 +0200
> From: "Jakob Hedegaard" <Jakob.Hedegaard at agrsci.dk>
> Subject: [BioC] Limma - using both array and spot weights in lmFit
> To: <bioconductor at stat.math.ethz.ch>
> Message-ID:
> 	<EA09C4B2B0F16E44B8F3311629493C0D0346E77C at DJFPOST01.djf.agrsci.dk>
> Content-Type: text/plain;	charset="iso-8859-1"
>
> Hi List
>
> I use spot weights (binary) to A) ignore spots "Not found" in GenePix during normalization:
>
> RG <- read.maimages(fnames, source="genepix", path=datadir, columns=columns,
> other.columns=other.columns, wt.fun=wtflags(weight=0,cutoff=0))
> MA <- normalizeWithinArrays(RG, method="loess", weights=RG$weights, bc.method="none")
>
> ... and B) include "cDNA" spots only in lmFit (ignoring various control spots):
>
> MA$weights[MA$genes$Status!="cDNA",] <- 0
> fit <- lmFit(MA, design, ndups=2, weights=MA$weights, method="robust", correlation=cor$cor)
>
>
> Does the use of array weights in lmFit...
>
> arrayw <- arrayWeights(MA, design, weights=MA$weights)
> fit2 <- lmFit(MA, design, ndups=2, weights=arrayw, method="robust", correlation=cor$cor)
>
> ... exclude the individual spot weights?

No. arrayw is just a vector of array weights.

> Are all spots from an "array weighted" array given the same weight in lmFit - thereby ignoring the
> "original" spot weights (in MA$weights)?

In your call to lmFit above, every spot on each array gets the same weight.

> How can the various control spots then be ignored in lmFit?

The way to remove control spots from the fit is simply to subset the MA object.  This is much
better than hacking the weight matrix.  See the Weaver case study (Section 11.5) in the Limma
User's Guide for an example.

Should you want to combine array and spot weights (as distinct from removing control probes), then
multiple the array weights by the spot weights, e.g.,

   library(statmod)
   W <- matvec(RG$weights, arrayw)
   lmFit(MA, design, weights=W)

Best wishes
Gordon

>> arrayw
>         1         2         3         4         5         6         7         8
> 0.9003237 0.5224861 0.8302243 0.7377452 0.6727376 1.1749278 1.0347314 0.4233758
>         9        10        11        12        13        14        15        16
> 0.5922037 0.5770843 1.1066639 0.5070331 1.1136625 0.8131061 0.9608490 0.3578452
> ............
>        81        82        83        84        85        86        87        88
> 1.1762249 0.8947206 1.0537712 1.2835124 1.4650973 1.8305352 1.4610934 1.4539068
>        89        90        91        92        93        94        95        96
> 1.3118581 1.1803798 1.5399506 1.2547482 1.3258992 1.0470026 1.0130834 1.1672451
>
>
>> sessionInfo()
> R version 2.4.1 (2006-12-18)
> x86_64-redhat-linux-gnu
>
> locale:
> LC_CTYPE=en_US.UTF-8;LC_NUMERIC=C;LC_TIME=en_US.UTF-8;LC_COLLATE=en_US.UTF-8;LC_MONETARY=en_US.UTF-8;LC_MESSAGES=en_US.UTF-8;LC_PAPER=en_US.UTF-8;LC_NAME=C;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_US.UTF-8;LC_IDENTIFICATION=C
>
> attached base packages:
> [1] "stats"     "graphics"  "grDevices" "utils"     "datasets"  "methods"
> [7] "base"
>
> other attached packages:
>     MASS  statmod      GDD    limma
> "7.2-30"  "1.3.0"  "0.1-8" "2.9.16"
>
>
>
> Looking forward hearing from you
>
> Jakob Hedegaard
> Project scientist
>
> UNIVERSITY OF AARHUS
> Faculty of Agricultural Sciences
> Dept. of Genetics and Biotechnology
> Blichers All? 20, P.O. BOX 50
> DK-8830 Tjele