[BioC] Detection of differential expression using limma

Sean Davis sdavis2 at mail.nih.gov
Mon Oct 13 18:14:59 CEST 2008


On Mon, Oct 13, 2008 at 11:40 AM, Christian Eisen
<mai at christianeisen.com> wrote:
> Hello to all,
>
> first of all I have to apologize for posting this several times, oviously
> the web interface of my institute mail adress doesn't let me post via
> mailing list...whatever...
> I am quite new to the topic of microarray analysis but I have a solid
> background
> on R. I am running a one-color Array from Agilent for my diploma thesis and
> I am currently analyzing the data I got.
> I always thought that, when looking at the raw data, if a gene gives a low
> value
> (intensity) within one considered group and a high value (value) within the
> other group considered
> this particular gene can be regarded to be upregulated respectively
> downregulated, whatever
> comparison of the two groups is interesting.
> However I was proven wrong when looking at preprocessed data of such two
> genes.
> The now transformed data for the two genes doesn't differ at all or very
> much.
> Sure I know it is log2 transformed and processed but however, going from a
> 10-fold change
> down to a 0,6-fold change seems a little bit far off for me.
> There can't be that much noise in such a signal...or am I wrong?
> As I said maybe my limited knowledge just doesn't let me interpret the data
> apropriately
> therfore I would be grateful for any kind of help.
>
> I tried VSN normalization, quantile normalization and a simple log2
> transformation of the raw data.
> For differential expression analysis I use linear models implemented in the
> limma package
> as described in the vignette for one-color data.
> Oddly enough these analysis always brings up genes which aren't
> differentially expressed at all
> between the two considered groups. Rather than that it lists genes as
> "significantly" differently expressed
> between the two considered groups which show a raw intensity data as well as
> normalized data fold-change of maximal 1.0.
>
> I really hope someone can help me clearify some of the issues mentioned
> above,since
> I really can't expect any help from my people here since they are even more
> unfamiliar than me.
> I appreciate any kind of input!
>
> Thanks a lot and best wishes
> Christian

Hi, Christian.  How are you calculating fold change?  When calculating
fold change on the log scale, you need to subtract and not divide.  If
that isn't the problem, then you will need to be a bit more explicit
about what you have done and what the numbers are.  Providing some
code, some sample output showing the problem, and your sessionInfo()
are all important pieces to getting useful help.

Sean



More information about the Bioconductor mailing list