[BioC] Missing values - metaArray, GeneMeta, RankProd, metahdep

mjonczyk at biol.uw.edu.pl mjonczyk at biol.uw.edu.pl
Fri Aug 12 00:25:56 CEST 2011

Hi John,

thanks for reply.
I've performed a part of example analysis from metaArray, I haven't
succeeded to perform analysis on a dataset from three partially
overlapping experiments. Zscore function gave results only for the
common part.

If I found a way to transform various experiments to effect-size
scale (probably with inverse-variance method as recommended in
Ramasamy et al. 2008 - very useful article) or something similar
I will try metahdep.

Best Regards,
Maciej Jończyk

> Hi Maciej,
> My metahdep package doesn't require all datasets to come from the same array
version -- so
while the package's tutorial vignette doesn't explicitly say anything about missing
values, the example there does allow for missing values in this sense (some
genes not
represented in all datasets).
> As to your second question, if you're combining studies with fundamentally
platforms (Affy for some, two-color for others, for example), you'll need to be
careful to
define an effect size (or some measure of differential expression) that means
the same
thing on all platforms, and for which you can calculate a meaningful standard
error.  I
think trans-platform cases like this lead to the appeal of alternative scale
methods such
as those in the metaArray package (which uses the 'probability of expression' or POE
> Regards,
> John Stevens
> ________________________________________
> From: bioconductor-bounces at r-project.org [bioconductor-bounces at r-project.org]
on behalf of
mjonczyk at biol.uw.edu.pl [mjonczyk at biol.uw.edu.pl]
> Sent: Tuesday, August 09, 2011 1:45 PM
> To: bioconductor at r-project.org
> Subject: [BioC] Missing values - metaArray, GeneMeta, RankProd, metahdep
> Dear List Members,
> I'd like to perform a metaanalysis for few (minimum three) datasets.
> Some of them are one channel (Affymetrix), some two-colour data.
> Obviously, when I join, say three studies I will have big dataset in
> which some data will be missing. Just because some probes will be present
> in two experiments and absent in third experiment. I'd prefer not to
> use *only* probes which are present on all arrays, but also this which
> are present on two arrays.
> My *question* is: which package (metaArray, GeneMeta, RankProd, metahdep)
> could handle metaanalysis with missing data - in tutorials/manuals it is
> not stated if missing data is allowed. Although metaArray cites MergeMaid
> package as a example software for merging data with non-overlapping genes.
> Second *question* which of the above package will best handle dataset
> from merged two-colour and affymetrix studies?
> I'd also be grateful for directions to additional tutorials, materials, etc.
> Best Regards,
> Maciej Jończyk
> Maciej Jończyk, MSc
> Department of Plant Molecular Ecophysiology
> Institute of Plant Experimental Biology
> Faculty of Biology, University of Warsaw
> 02-096 Warszawa, Miecznikowa

More information about the Bioconductor mailing list