[BioC] RMA t-test

[Rafael A. Irizarry]

hi! fyi, all the ROC curves in the NAR paper comparing expression measures
was done on 1-1 comps (i.e. 2 chips). rma seems to perform weel
according to these curves. notice that if you only have one
chip median polish turns into a median which in my opinion is not  a
terrible thing to use. 

as for t-tests with two arrays all you could use for estimating
variability (denominator in the t-test) are the residuals from the median
polish fit. these have (at least) two problemsB
1) as james points out they have no information about biological
variation
2) its not clear what the statistical properites of these residuals are.
the se estimate one gets with expresso-rma are given only as an ad-hoc
estimate of uncertainty of the expression estimates due to technology.

hope this helps,
rafael
On Mon, 28 Jul 2003, James MacDonald wrote:

> I would be cautious about using RMA with only two chips. You will be
> estimating the probe-specific intensity with only two observations, so
> it is doubtful that the estimate will be very accurate. I recall reading
> somewhere that a good minimum number of chips is around 5-6 for RMA.
> 
> As for a t-test with only one observation per group, this is not
> possible. How are you going to estimate the variance for each group?
> Without replication all you can do is ratios, and you are then stuck
> with the assumption that large ratios equal significant differences.
> 
> Jim
> 
> 
> 
> 
> James W. MacDonald
> Affymetrix and cDNA Microarray Core
> University of Michigan Cancer Center
> 1500 E. Medical Center Drive
> 7410 CCGC
> Ann Arbor MI 48109
> 734-647-5623
> 
> >>> DMITRY GRIGORYEV <dgrigor1 at jhmi.edu> 07/28/03 12:08PM >>>
> Hi everyone.
> 
> One quick question.
> When I run RMA on just two chips, how could I conduct pairwise t-test
> for each probe set between these chips?
> 
> Thank you 
> 
> Dima
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch 
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>