On 12/29/05 6:45 AM, "Ankit Pal" <pal_ankit2000 at="" yahoo.com=""> wrote:
> Hello,
> Could anyone tell me how to go about doing an analysis for a
single
> microarray slide using limma.
You can't is the short answer. Limma employs a model that assumes
microarray replication is present.
> Below is the code I used to specify the design,
>
> fit <- lmFit(MA, design=c(1))
>
> But I get the following errorr once I go to fit <- eBayes(fit)
>
> Error in ebayes(fit = fit, proportion = proportion, stdev.coef.lim
=
> stdev.coef.lim) :
> No residual degrees of freedom in linear model fits
>
> I am not a statistician, so I need help to interpret the above
error.
In this case, it means that you are trying to do a t-test with only
one
sample--you can't.
The simplest way to go about this is to rank the genes by fold-change
(two-channel data). That is really the best you can do with only one
slide.
Determining statistical significance is another question (and people
have
tried to answer it), but I would argue that doing so really isn't that
meaningful and that if you really want to know what is "significant",
you
need some replicates (the number of which depends on the experimental
conditions and design).
Hope that clarifies things a bit.
Sean
A basic principle of statistical analysis of differential expression
is to compare differences between conditions to differences among
replicates within condition.
If you have no replication, you cannot use a statistical method such
as LIMMA, MAANOVA, t-tests, Wilcoxon test or SAM.
All you can do is order the differences (M) from largest to smallest,
but this does not tell you anything about statistical significance.
--Naomi
At 06:45 AM 12/29/2005, Ankit Pal wrote:
>Hello,
> Could anyone tell me how to go about doing an analysis for a
> single microarray slide using limma.
> Below is the code I used to specify the design,
>
> fit <- lmFit(MA, design=c(1))
>
> But I get the following errorr once I go to fit <- eBayes(fit)
>
> Error in ebayes(fit = fit, proportion = proportion,
> stdev.coef.lim = stdev.coef.lim) :
> No residual degrees of freedom in linear model fits
>
> I am not a statistician, so I need help to interpret the above
error.
>
> Thanks and regards
>
> Ankit
>
>
>
>
>
>---------------------------------
>
>
> [[alternative HTML version deleted]]
>
>_______________________________________________
>Bioconductor mailing list
>Bioconductor at stat.math.ethz.ch
>https://stat.ethz.ch/mailman/listinfo/bioconductor
Naomi S. Altman 814-865-3791 (voice)
Associate Professor
Dept. of Statistics 814-863-7114 (fax)
Penn State University 814-865-1348
(Statistics)
University Park, PA 16802-2111
> A basic principle of statistical analysis of differential expression
> is to compare differences between conditions to differences among
> replicates within condition.
> If you have no replication, you cannot use a statistical method such
>
> All you can do is order the differences (M) from largest to
smallest,
> but this does not tell you anything about statistical significance.
there might be replication of material ON the slide.
in this case, one should be able to use limma to do the
within-slide normalization, and then some analysis that
makes use of the variability among replicate measures
to obtain something more informative than a raw ranking.
the original poster will likely have to contact
a statistician with some R expertise to carry this out.
but it is a case near the boundary that is worth being
able to handle.
Within slide replication is NOT suitable for statistical testing. It
measures only technical variation within the slide, which is usually
much smaller than any other source of variation.
That is why in limma we use Blocks to handle replicate spots.
--Naomi
At 10:47 AM 12/30/2005, Vincent Carey 525-2265 wrote:
> > A basic principle of statistical analysis of differential
expression
> > is to compare differences between conditions to differences among
> > replicates within condition.
> > If you have no replication, you cannot use a statistical method
such
> >
> > All you can do is order the differences (M) from largest to
smallest,
> > but this does not tell you anything about statistical
significance.
>
>there might be replication of material ON the slide.
>in this case, one should be able to use limma to do the
>within-slide normalization, and then some analysis that
>makes use of the variability among replicate measures
>to obtain something more informative than a raw ranking.
>
>the original poster will likely have to contact
>a statistician with some R expertise to carry this out.
>but it is a case near the boundary that is worth being
>able to handle.
>
>_______________________________________________
>Bioconductor mailing list
>Bioconductor at stat.math.ethz.ch
>https://stat.ethz.ch/mailman/listinfo/bioconductor
Naomi S. Altman 814-865-3791 (voice)
Associate Professor
Dept. of Statistics 814-863-7114 (fax)
Penn State University 814-865-1348
(Statistics)
University Park, PA 16802-2111
