[BioC] MetaArray - results - how to interpret

[Adrian Johnson]

Hi Rob,
Thanks for your quick reply.


Is there a way to select again from the list of 500 genes?

Here is what I am trying to achieve:

1. I have 8 different Affy studies for Lung.
2. I have list of 200 genes related to a biological process. Some of
the 200 genes are differentially expressed (using Limma) between lung
cancer and normal lung samples across studies but not all of them are
significant in every study.     Visualizing these 200 genes on a heat
map, it appears majority of them over-expressed in tumors compared to
normals (although they are over-expressed they are not significant at
p-value 0.001) (red block in heat-map with red for up and green for
down expression).

Irrespective whether some genes are significant or not significant at
a given p-value in individual studies,  I want group of genes say 140
out of 200 are significantly differentially expressed across studies.

For instance below:

Study 1       40/200 are differentially expressed (limma p.value < 0.001)
Study 2       90/200 are differentially expressed       "
Study 3       80/200 are differentially expressed.      "

What I did is - I took union of all the differentially expressed genes
(that are 200) and want to analyze using poe.mcmc model.

My aim is to select most significant differentially expressed genes
across all 8 studies. My initial matrix after merge and intersection
is 200 genes by 190 samples (tumor and normal from 8 different
studies) for poe.mcmc

coming back to test dataset in metaarray package, where 500 genes are
being tested, my question is:

How do I select top differentially expressed genes from this 500 genes
from poe.mcmc$poe matrix.  Can I select genes based on any rank or
p-value again from poe.mcmc$poe matrix?
I am assuming not all 500 genes are significant after mcmc and poe
transformation.
Please pardon me if I am sounding too naive.

Thanks
Adrian




On Mon, Oct 22, 2012 at 1:22 PM, Robert Scharpf <rscharpf at jhsph.edu> wrote:
> Adrian,
>
> You probably want poeRes$poe.  'poe' is short for "probability of expression" and is a transformed matrix of gene expression values (number of genes x number of samples).  Interpretation of under and over-expression depends on how the phenotype is defined.  According to the poe.mcmc helpfile, if normal is group 1 and is coded as '1' and group 2 is coded as '0', then positive values on the poe scale would be interpreted as the probability that the gene is over-expressed in group 2 relative to group 1.
>
> POE for biologists:
>
> http://www.biotechniques.com/multimedia/archive/00072/Mar03Scharpf_72034a.pdf
>
> Since you have 3 datasets, one option is to run poe.mcmc on the three datasets independently and use ordinary measures of differential expression on the combined studies (I believe Shen et al., 2004 BMC Genomics describes this appoach).  fyi, other packages useful for analyzing multiple studies include the R packages RankProd (uses a rank product), XDE (a Bayesian multilevel model; Scharpf et al., 2009 JASA ), and the references therein.
>
> Rob
>
>
> On Oct 22, 2012, at 11:58 AM, Adrian Johnson <oriolebaltimore at gmail.com> wrote:
>
>> Dear group,
>> Pardon me for re-post.
>>
>> I am writing to seek some help in interpreting MetaArray poe.mcmc results.
>>
>>
>> After running poe.mcmc, the resulting results object is a complicated
>> result ( I have biology training and minimal statistics).
>>
>> I am trying to extract those genes that are consistently
>> differentially expressed (over-expressed in condition 1 - metastasis)
>> across all 3 datasets given in test data.
>>
>> The result object poeRes has following names
>>> names(poeRes)
>> [1] "alpha"        "mug"          "kappaposg"    "kappanegg"    "sigmag"
>> [6] "piposg"       "pinegg"       "mu"           "tausqinv"     "gamma"
>> [11] "lambda"       "pil.pos.mean" "pil.pos.prec" "pil.neg.mean" "pil.neg.prec"
>> [16] "kap.pos.rate" "kap.neg.rate" "poe"          "accept"
>>
>>
>> How do I choose those genes that are over or under-expressed in
>> metastatic tumors compared to normals.  I have 0 in accept.
>>
>> I do not know which object (alpha, mug, kappa pos and neg, pi pos and
>> neg, mu, tau, gamma, lambda etc..) has the result to pick from.
>>
>> The vignette does not have additional details on interpretation.
>> Could Drs. Choi or Ghosh, please help.
>>
>> Thanks
>> Adrian.
>>
>> _______________________________________________
>> Bioconductor mailing list
>> Bioconductor at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor
>