[BioC] Analyzing expression Affymetrix Hugene1.0.st array

[James W. MacDonald]

On 9/28/2012 11:41 AM, Steve Lianoglou wrote:
> Hi,
>
> Totally agree w/ everything Jim has said (this is usually a smart
> thing to do), but just wanted to comment on:
> On Fri, Sep 28, 2012 at 11:21 AM, James W. MacDonald<jmacdon at uw.edu>  wrote:
> [snip]
>
>> I think this becomes a bit more difficult with the Gene ST arrays, as the
>> negative controls have a nasty habit of looking not only expressed, but
>> differentially expressed. A lot of these controls are supposed to target
>> introns, which makes me wonder how much of the total RNA extracted from a
>> cell is mRNA for which the introns have yet to be excised.
> Instead of such negative control probes, perhaps you (Juan) might know
> something about the types of cells you have data from. In particular,
> perhaps you can justify identifying a multitude of genes that you know
> not to be expressed in these cell types and use some statistics over
> their probe expression to rig up a lower bound of your detection
> limit.
>
> If you don't know this info, and have no expert to ask, maybe you can
> find rna-seq data in cells "close" (using some definition of "close"
> that makes you comfortable) and use that to find such non-transcribed
> genes.
>
> I guess there's also going to probe (sequence content) effects that
> affect the expression readout of these "silent" probes and what not,
> but ... if you're going for some heuristic thing that you're not using
> as the lynchpin of your study, then perhaps this is passable.

And this in a nutshell is why I am uncomfortable with trying to discern 
expressed from not expressed genes using microarray data. There are any 
number of variables that affect the fluorescence of a given spot on a 
microarray, only one of which is the binding of the target cDNA, and 
even the binding of that cDNA is predicated on GC content, and various 
other stoichiometric variables that I don't think we know or can appreciate.

So I agree with Steve here, that if this is some tangential aspect of 
the study maybe it is OK. But if it is the main thrust of the analysis, 
then caveat emptor.

Best,

Jim


>
> -steve
>

-- 
James W. MacDonald, M.S.
Biostatistician
University of Washington
Environmental and Occupational Health Sciences
4225 Roosevelt Way NE, # 100
Seattle WA 98105-6099