[BioC] se.exprs/signficance analysis

[Rafael A. Irizarry]

the values returned by se.exprs from expresso with rma options is the
following: for each probeset obtain the residuals from the median polish
procedure. take the MAD of these and multiply by 1.483 (if i remeber
correctly this constant makes MAD an unbiased estimate of variance
for normal data). to get se, divide by sqrt(#probes). notice its a
very ad-hoc estimate of the se of the expression estimate. 

notice this se estimate tries to account for the uncertanty due to
technology measuremnent error. it says nothing about biological variance.
so your hypothesis tests are contrained to differences between the
specimens for which you have RNA, you can not extrapolate to populations.
if this is ok, then you could consider the estimate (log expression) / se
a t-test and get p-values. however, given the ad-hoc nature of the se
estimate and the
fact that the median polish esimtate is not a mean (central limit theorem
doesnt apply) i would be very careful interpreting these. a more sensible
approach would be to use rlm instead of median polish.



On Mon, 5 May 2003, Ann Hess wrote:

> 
> Is there any documentation available for the values returned by se.exprs?
> 
> Is there a way to perform a significance analysis WITHOUT duplicate
> arrays?  For example, if you have two arrays for each of two treatments
> you can use multtest to identify differentially expressed genes.  Is there
> a way to determine significance with only one array for each treatment?
> 
> Thanks.
> 
> Ann
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>