Hi all,
Although limma-voom seems to give us really good results in our RNA-
seq experiments, we still have some doubt about y$E values (i.e.
expression values after voom). looking at function implementation it
seems that voom performs quantile normalization using
normalizeBetweenArrays function (default option). I need a
clarification here: it has been explained many times that limma (and
edgeR too) do not need to take into account gene length as they
perform gene-wise statistical tests: feature length does not change
and so it is good to take into account library size. It has also been
explained that this method is good for DGE but does not necessarily
works for "absolute quantification" of a transcript. But. AFAIK (but I
may be wrong) quantile normalization works by considering all features
in each expression quantile together; in the same quantile there could
be low expressed long genes and highly expressed short genes (if we
only consider CPM), and I'm missing the rationale behind. If the
normalization process is correct I should be able to use y$E values as
"absolute" quantification (or absolute-like) of a transcript. If the
y$E cannot be used as absolute quantification, is quantile
normalization the proper way to do this?
Sorry for the confused mail, it's early morning and I really need some
sleep...
Thanks
d
Dear Davide,
It is better to read the documentation for voom instead of trying to
interpret the code. You would then see that the default normalization
is
"normalization.method=none".
In fact, voom uses the normalization factors computed by edgeR and
placed
in the DGEList object to compute effective library sizes, so by
default
voom's normalization is exactly as for edgeR.
Best wishes
Gordon
> Date: Tue, 5 Feb 2013 07:07:51 +0100
> From: Davide Cittaro <cittaro.davide at="" hsr.it="">
> To: Bioconductor mailing list <bioconductor at="" r-project.org="">
> Subject: [BioC] limma-voom and y$E
>
> Hi all,
>
> Although limma-voom seems to give us really good results in our RNA-
seq
> experiments, we still have some doubt about y$E values (i.e.
expression
> values after voom). looking at function implementation it seems that
> voom performs quantile normalization using normalizeBetweenArrays
> function (default option). I need a clarification here: it has been
> explained many times that limma (and edgeR too) do not need to take
into
> account gene length as they perform gene-wise statistical tests:
feature
> length does not change and so it is good to take into account
library
> size. It has also been explained that this method is good for DGE
but
> does not necessarily works for "absolute quantification" of a
> transcript. But. AFAIK (but I may be wrong) quantile normalization
works
> by considering all features in each expression quantile together; in
the
> same quantile there could be low expressed long genes and highly
> expressed short genes (if we only consider CPM), and I'm missing the
> rationale behind. If the normalization process is correct I should
be
> able to use y$E values as "absolute" quantification (or absolute-
like)
> of a transcript. If the y$E cannot be used as absolute
quantification,
> is quantile normalization the proper way to do this?
>
> Sorry for the confused mail, it's early morning and I really need
some
> sleep...
>
> Thanks
>
> d
>
______________________________________________________________________
The information in this email is confidential and
intend...{{dropped:4}}
