Hi all,
Firstly thanks for all the help in the past!
Now, I have biological replicate arrays for which different amounts of
RNA have been used (3 micrograms and 4 micrograms). So a fairly big
difference of over 30%.
I am performing normalizeWithinArrays and then normalizeBetweenArrays
in
limma. The between array normalization is necessary since the arrays
were scanned at different PMT settings (both in the linear scale) but
I
also need to normalize for the different amounts of RNA if this is
possible. I imagine that the relationship between the amount of RNA
added to a slide and the amount that hybridizes to the array is not a
linear relationship, probably sigmoidal but I haven't tested this. If
this is so, would normalizeBetweenArrays account for this? Is there a
different type of normalization that would? Or could I transform the
data in some way so that it would work for both different RNA amounts
and different scanning settings?
Also is it possible to visualize the normalized R and G values (but
not
as M and A values)? Since when I look at my top table I'm seeing genes
that appear yellowish on the arrays with similar values between arrays
(ok so these are raw values) and not the ones that appear bright red
or
green since the raw values are so different between them but the
ratios
are similar (eg Cy5=1000, Cy3=500 on array1 and Cy5=2000, Cy3=1000 on
array2). This makes me suspect this normalization is not enough for my
data.
I'm using R v1.8.1 and limma v1.6.1
Does anyone have any suggestions or comments please?
Thanks,
Helen
[[alternative HTML version deleted]]
Hi,
I have encountered a problem with the pamr.from.excel function.
When I try to perform this function on my file it seems that I have
781
genes whereas my files has 1000 genes. When I compare the geneid
obtained and those who are present in my file I make the remark that
some packages of genes have benn removed. I mean that I have the 100
first genes and then the 20 following genes are removed and so on.....
I
really don't understand what happened !
For details I had :
data<-pamr.from.excel("file.txt",90,sample.labels=TRUE,batch.labels=FA
LSE)
Read 70470 items
Read in 781 genes
Read in 88 samples
Read in 88 sample labels
Make sure these figures are correct!!
Could you help me ?
Thank you,
Regards
Willy
Helen
I'll try and tackle these issues, but perhaps an e-mail to the
microarray-norm@ebi.ac.uk mailing list would also help
>but I also need to normalize for the different amounts of RNA if this
is possible.
Most normalisation procedures assume that most genes are not changing
and therefore that the "average" log(ratio) is zero. If this
assumption holds for your data, then any normalisation procedure which
sets the average log(ratio) to zero (median, loess etc) *should* also
be handling the different amounts of RNA (but not in a very
sophisticated manner, see below)
>I imagine that the relationship between the amount of RNA added to a
slide and the
>amount that hybridizes to the array is not a linear relationship,
probably sigmoidal
>but I haven't tested this.
I teach on the Birmingham Microarray Technology Course and data from
this suggests that there is a sigmoidal relationship between
concentration of DNA on the spot and intensity. I imagine the same
holds true for amounts of RNA
>If this is so, would normalizeBetweenArrays account for this?
Median certainly won't, as this undoubtedly assumes a linear
relationship. However, I think Loess should account in some way for
the sigmoidal relationship we assume is present between amount of RNA
and intensity.
>Is there a different type of normalization that would?
Not unless you have done previous experiments to define the
relationship between amount of RNA and intensity or ratio on your
system. There may be some way of doing this if you have spiked in
controls, but I am not sure how.
>Also is it possible to visualize the normalized R and G values (but
not as
>M and A values)?
Someone has definitely posted to this list before about accessing
normalised R and G values, so it is possible, but I can't remember how
I'm not helping am I? ;-)
Mick
Here is a function to convert M and A values back to normalized R and
G
values.
convert.back <- function(M, A){ #Here M = maM values and A = maA
values
G <- (2*A-M)/2
R <- (2*A+M)/2
return(cbind(R,G))
}
HTH,
Jim
James W. MacDonald
Affymetrix and cDNA Microarray Core
University of Michigan Cancer Center
1500 E. Medical Center Drive
7410 CCGC
Ann Arbor MI 48109
734-647-5623
>>> "michael watson (IAH-C)" <michael.watson@bbsrc.ac.uk> 05/11/04
06:38AM >>>
Helen
I'll try and tackle these issues, but perhaps an e-mail to the
microarray-norm@ebi.ac.uk mailing list would also help
>but I also need to normalize for the different amounts of RNA if this
is possible.
Most normalisation procedures assume that most genes are not changing
and therefore that the "average" log(ratio) is zero. If this
assumption
holds for your data, then any normalisation procedure which sets the
average log(ratio) to zero (median, loess etc) *should* also be
handling
the different amounts of RNA (but not in a very sophisticated manner,
see below)
>I imagine that the relationship between the amount of RNA added to a
slide and the
>amount that hybridizes to the array is not a linear relationship,
probably sigmoidal
>but I haven't tested this.
I teach on the Birmingham Microarray Technology Course and data from
this suggests that there is a sigmoidal relationship between
concentration of DNA on the spot and intensity. I imagine the same
holds true for amounts of RNA
>If this is so, would normalizeBetweenArrays account for this?
Median certainly won't, as this undoubtedly assumes a linear
relationship. However, I think Loess should account in some way for
the
sigmoidal relationship we assume is present between amount of RNA and
intensity.
>Is there a different type of normalization that would?
Not unless you have done previous experiments to define the
relationship between amount of RNA and intensity or ratio on your
system. There may be some way of doing this if you have spiked in
controls, but I am not sure how.
>Also is it possible to visualize the normalized R and G values (but
not as
>M and A values)?
Someone has definitely posted to this list before about accessing
normalised R and G values, so it is possible, but I can't remember how
I'm not helping am I? ;-)
Mick
_______________________________________________
Bioconductor mailing list
Bioconductor@stat.math.ethz.ch
https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
There is already a function in Limma to do this
MA.RG() and RG.MA() works both ways MA -> RG and RG ->MA
Natalie
On Tue, 11 May 2004, James MacDonald wrote:
> Here is a function to convert M and A values back to normalized R
and G
> values.
>
> convert.back <- function(M, A){ #Here M = maM values and A = maA
> values
> G <- (2*A-M)/2
> R <- (2*A+M)/2
> return(cbind(R,G))
> }
>
> HTH,
>
> Jim
>
>
>
> James W. MacDonald
> Affymetrix and cDNA Microarray Core
> University of Michigan Cancer Center
> 1500 E. Medical Center Drive
> 7410 CCGC
> Ann Arbor MI 48109
> 734-647-5623
>
> >>> "michael watson (IAH-C)" <michael.watson@bbsrc.ac.uk> 05/11/04
> 06:38AM >>>
> Helen
>
> I'll try and tackle these issues, but perhaps an e-mail to the
> microarray-norm@ebi.ac.uk mailing list would also help
>
> >but I also need to normalize for the different amounts of RNA if
this
> is possible.
>
> Most normalisation procedures assume that most genes are not
changing
> and therefore that the "average" log(ratio) is zero. If this
assumption
> holds for your data, then any normalisation procedure which sets the
> average log(ratio) to zero (median, loess etc) *should* also be
handling
> the different amounts of RNA (but not in a very sophisticated
manner,
> see below)
>
> >I imagine that the relationship between the amount of RNA added to
a
> slide and the
> >amount that hybridizes to the array is not a linear relationship,
> probably sigmoidal
> >but I haven't tested this.
>
> I teach on the Birmingham Microarray Technology Course and data from
> this suggests that there is a sigmoidal relationship between
> concentration of DNA on the spot and intensity. I imagine the same
> holds true for amounts of RNA
>
> >If this is so, would normalizeBetweenArrays account for this?
>
> Median certainly won't, as this undoubtedly assumes a linear
> relationship. However, I think Loess should account in some way for
the
> sigmoidal relationship we assume is present between amount of RNA
and
> intensity.
>
> >Is there a different type of normalization that would?
>
> Not unless you have done previous experiments to define the
> relationship between amount of RNA and intensity or ratio on your
> system. There may be some way of doing this if you have spiked in
> controls, but I am not sure how.
>
> >Also is it possible to visualize the normalized R and G values (but
> not as
> >M and A values)?
>
> Someone has definitely posted to this list before about accessing
> normalised R and G values, so it is possible, but I can't remember
how
>
> I'm not helping am I? ;-)
>
> Mick
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>
--
______________________________
Natalie Thorne
Research Associate
Computational Biology Group
Hutchison/MRC Research Centre
Department of Oncology
University of Cambridge
Hills Rd, Cambridge CB2 2XZ
Email: npt22@cam.ac.uk
Phone: +44 (0)1223 763389
Fax : +44 (0)1223 763262
Here is my go at it:
1. The biggest worry is that in the more intense sample there will be
more
saturation. The level of differential expression cannot be determined
from
the saturated spots.
2. If differential expression is really a ratio, then the amount of
RNA in
the sample should not affect measures of differential expression -
i.e.
(10R/10G)=R/G. Clearly the ratio idea is not exact, but taking
logarithms
before doing the analysis is based on this idea.
3. If you are using ANOVA or limma to do the computations, if (2) is
correct and if you have an array effect in your model, your measures
of
differential expression should be OK. Alternatively, you could put
"RNA
amount" as a factor in the model and you should be OK.
4. I agree that you should normalize between arrays. I am not very
familiar with the limma routines - you will want to normalize both the
mean
and the variance. Normalizing the variance should help reduce effects
due
to the amount of RNA in the sample.
5. If you are using an analysis that does not involve taking the
difference in logarithms, points 2-4 will not help you. In that case,
you
need to know the relationship between amount in the sample and
hybridization intensity.
--Naomi
At 05:04 PM 5/10/2004, Helen Cattan wrote:
>
>Hi all,
>
>Firstly thanks for all the help in the past!
>
>Now, I have biological replicate arrays for which different amounts
of
>RNA have been used (3 micrograms and 4 micrograms). So a fairly big
>difference of over 30%.
>
>I am performing normalizeWithinArrays and then normalizeBetweenArrays
in
>limma. The between array normalization is necessary since the arrays
>were scanned at different PMT settings (both in the linear scale) but
I
>also need to normalize for the different amounts of RNA if this is
>possible. I imagine that the relationship between the amount of RNA
>added to a slide and the amount that hybridizes to the array is not a
>linear relationship, probably sigmoidal but I haven't tested this. If
>this is so, would normalizeBetweenArrays account for this? Is there a
>different type of normalization that would? Or could I transform the
>data in some way so that it would work for both different RNA amounts
>and different scanning settings?
>
>
>
>
> [[alternative HTML version deleted]]
>
>_______________________________________________
>Bioconductor mailing list
>Bioconductor@stat.math.ethz.ch
>https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
Naomi S. Altman 814-865-3791 (voice)
Associate Professor
Bioinformatics Consulting Center
Dept. of Statistics 814-863-7114 (fax)
Penn State University 814-865-1348
(Statistics)
University Park, PA 16802-2111
