[BioC] interaction analysis

[Robert Castelo]

dear Phillippe,

i might be a little late to help you out, but just in case, you may have
found a message of mine prior to this one where i'm announcing a new
package for Bioconductor called 'qpgraph'. This package *does not*
address what you are asking for, namely to identify interaction terms in
a transcriptional regulatory network from microarray data, since the
package assumes that employed microarray data set forms a multivariate
normal sample which implies assuming linearity and additivity. These two
assumptions break down in the presence of interacting terms.

however, i think it could help you to reduce the number of candidate
subsets of TFs where you would try the approach you're proposing or any
other.

the reason why i think this is because if we assume the following model
for transcriptional regulation of a target by two interacting TFs:

TG = beta1 x TF1 + beta2 x TF2 + beta12 x TF1 x TF2

it implies that next to the marginal effect of the interaction term on
TG there are marginal linear effects from the "main" terms (beta1 x TF1
+ beta2 x TF2) and, depending on the beta1, beta2 and beta12, these
marginal linear effects could be detected with a methodology assuming a
linear model like the one of qpgraph.

so, i would propose you to use qpgraph to define subsets of two or more
TFs interacting with target genes (you may want to further filter those
with binding data), and then use those subsets in a further analysis
like the one you suggest.

best wishes,

robert.


On Fri, 2009-01-09 at 15:29 -0500, phguardiol at aol.com wrote:
> Dear R users
> 
> I would like to have your opinion about the way (and the package that 
> would be able to run this issue) to analyse gene expression data 
> "around transcription factors":
> there are biological situations in which multiple factors 
> (Transcription factors TF or other cofactors) are involved in the fact 
> that a gene is transcribed, in a way that if TF1 is expressed but not 
> TF2, gene3 is lowly expressed, same if TF2 alone is expressed, and it 
> is only when both TF1 & TF2 are expressed that gene3 is highly 
> expressed. This leads me to think about synergy between TF / 
> interaction term.
> I d like to explore this in a model we work on in our lab using GEX 
> chips. I was thinking of creating a new variable that would be TF1 
> signal intensity x TF2 signal intensity and see which gene transcripts 
> correlate with this product term. What is your opinion about such an 
> approach ? Is there a package that would allow me to run this, given 
> the fact that I would like to study all 2x2 interaction terms with the 
> whole set of TF identified on our HT12 illumina chips.
> 
> As anyone being doing this kind of analysis ? If so should I consider 
> using raw signal of log transformed product term ?
> 
> Hoping to be not too confused
> 
> Best wishes for 2009
> 
> Regards
> 
> Philippe Guardiola
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor
>