[BioC] AffyPLM

[Francois Collin]

Paul,
The weights are derived from the model fit residuals -
they are transformations of the residuals standardized
by the model fit residual variance (fit here being
probe set specific).  Weights will be 1.0 if the
residuals are small compared to the residual variance,
and then decrease toward zero as the value of the
absolute standardized residuals increase.  On the chip
speudo-image of the weights, what is highlighted is
the spatial distribution of probes with large absolute
standardized residuals - outliers, in a sense, that
might have an impact on the fit, although this impact
is minimized by the robustness of the fit.  If there
is a local artifact - a scratch, uneven hybridization,
incomplete wash, bubble, etc - this will appear as a
cluster on the chip weight pseudo-image.  You could
also see a chip where residuals are uniformly elevated
throughout the chip  indicating that either the RNA
preparation, or the hybridization assay failed.

The pseudo images of the residuals are just images of
untransformed residuals.  Here you may see local
clusters that do not appear in the weights -
corresponding to slightly dim or slightly bright spots
which lead to elevated residuals, but not elevated
enough to be picked up by the weights.  These are
useful to detect effects which may be good to know but
are too subtle to be picked up in the weights.  In
general, the images of the residuals tell the same
story as the weights.

-francois


--- Lawrence Paul Petalidis <lpp22 at cam.ac.uk> wrote:
> Hello All,
> I am quite new to BioC and would appreciate your
> help on this. I am 
> experimenting with AffyPLM and taking a look the
> post AffyPLM quality 
> diagnostic pseudo-chip images [pset <- fitPLM(eset)
> and then 
> image(pset)   ].  Can anybody recommend how one
> shoud interpret these 
> images as a quality-checking step? Further, could
> one clarify for me 
> what the differences between the plots that show
> weight, and those that 
> show residuals?
> I thank you for your kind attention, Lawrence
> 
> 
> 
> ___________________________________
> Lawrence-Paul Petalidis
> University of Cambridge
> Department of Pathology
> Division of Molecular Histopathology
> Addenbrookes Hospital, Level 3, Lab Block
> Hills Road, CB2 2QQ
> Cambridge
> 
> Tel. : ++44 1223 762084
> Fax : ++44 1223 586670
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
>
https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor