[BioC] External expression data/ MAS5 Signal

[Sek Won Kong, M.D.]

Dear Laurent

i used log transformed and untransformed for both algorithms. and you are
right, the r square is good enough for those algorithms. however, the small
difference made quite a diffrerence when trying sam, samroc procedure for
choosing the significant genes. that's the problem. usually 50-60% genes
were found significant in both rma, li.wong, and MAS 5 with given same
threshold or highest 500 rank genes.

thanks!
SW


----- Original Message -----
From: "Rafael A. Irizarry" <ririzarr@jhsph.edu>
To: "Laurent Gautier" <>
Cc: "Sek Won Kong, M.D." <skong@caregroup.harvard.edu>; <>
Sent: Monday, June 17, 2002 9:31 AM
Subject: Re: [BioC] External expression data/ MAS5 Signal


> also keep in mind that rma return log2(expression). if you want to make
> a direct comp you should look at 2^exprs(express(Plob))
>
> rma and li and dchip with no pm only give similar results. i think the
> average correlation is about .9 (on the log scale). for mas 5.0 its around
.8.
>
> On
> Mon, 17 Jun 2002, Laurent Gautier wrote:
>
> > On Thu, Jun 13, 2002 at 02:04:50PM -0400, Sek Won Kong, M.D. wrote:
> > (...)
> > >
> > > I wondered there's quite a difference in calculated expression values
from
> > > AvgDiff, AvgDiff2, li.wong and rma. Previous I used invariant set
normalized
> > > MAS 5 signal values for analysis. However when I tried to implement
the
> > > quantile normalization and rma, there's big differences in reults ( I
mean
> > > the final results the list of genes ). So I wondered the why, and
wanted to
> > > compare 5 different kinds of expression algorithms.
> >
> > The 'because' that answers the 'why' is probably linked with the fact
that
> > the methods are fairly different (see their respective description in
the
> > help pages)...
> > You may want to look at the individual probe sets for the genes for
which
> > the results are very different. (try the function get.PPset)... do not
> > hesitate to give us feedback about the differences observed.
> >
> > >
> > > And could you let me know how to calculate the PM-only model of
Li-Wong in
> > > affy? li.wong seems like PM-MM model.
> > >
> >
> > In the package, we consider that PMs and MMs probe intensities are used
> > to compute 'adjusted' probe intensities. In the case of PMs-MMs,
> > it can be thought of subtracting the noise to the signal.
> >
> > The parameter 'bg' for express or 'bg.correct' for generateExprSet
> > (for Plob and Cel.container respectively) let you specify how to
> > get the 'adjusted' values.
> > The function 'bg.correct.pmonly' returns only the PMs and can be
> > used...
> >
> >
> > regards,
> >
> >
> > Laurent
> >
> >
> >
> >