[BioC] interpretation of vsn normalized data

[Adaikalavan Ramasamy]

I agree. There is a difference between hypothesis testing and
estimation. Microarray is good for hypothesis testing (and hypothesis
generating) but not good at estimation.

Consider the following two observations :

1) Different platforms and softwares (e.g. preprocessing algorithm) give
rise to different dynamic ranges and there is no agreed gold standard. 

2) It is common that biologist validate the genes ranked highly by
microarray results with a method such as real time PCR to estimate the
fold change accurately.


Given these two observations I do not see why biologists are obsessed by
the need to observe a two-fold change by microarrays !

Therefore whether the log is to the base 2, 10 or natural is irrelevant.

Regards, Adai



On Thu, 2006-02-16 at 18:16 +0000, Wolfgang Huber wrote:
> Hi Maurice,
> 
> in statistics it is sometimes useful to differentiate between (a) the 
> estimator and (b) the true underlying quantity that you want to estimate.
> 
> For example, if you want to estimate the expectation value of a 
> symmetric distribution, you can use the mean, or the median as 
> estimators. They are both correct, but depending on the data they can 
> provide different, and more or less appropriate answers.
> 
> With microarrays, (b) is the fold-change, that is the change in mRNA 
> abundance. The log-ratio of fluorescence intensities is a simple and 
> intuitive estimator for this, but if the fluorescence intensities become 
> small, this estimator can have unpleasant properties, like large 
> variance. The glog-ratio (what vsn provides) is an alternative 
> estimator, which avoids the variance explosion, for the price of being 
> biased towards 0 when the fluorescence intensities are small.
> 
> Note that the vsn function returns glog to base e (so a glog-ratio of 1 
> corresponds to an estimated fold change of exp(1) = 2.718..) while many 
> other packages use log2.
> 
>   Hope this helps
>   Wolfgang
> 
> 
> 
> 
> Maurice Melancon wrote:
> > Hello All,
> > 
> > I used vsn to normalze my one-channel cDNA microarray experiment.  I'm sorry
> > if this is an elementary question (I'm not a math person) but can vsn data
> > be interpreted in similar fashion to log2 data, e.g. 1 log vale = 2-fold
> > induction?  What would be the appropriate transformation to get to either
> > log2 or raw data from vsn data?
> > 
> > Briefly, what I did was to normalize using vsn, then I used SAS to run
> > anovas with pairwise comparisons and anova slicing.  Using the estimate
> > function returns estimated differences between the reported means.  I am
> > seeking then to bridge the gap between these estimates and actual fold
> > changes.  I think this can be done, but I am unsure about how to either
> > reverse-transform the vsn data or how to interpret it biologically (e.g. 1
> > log = 2x fold change)
> > 
> > WIth thanks
> > 
> > Maurice
> > 
> > 	[[alternative HTML version deleted]]
> > 
> > _______________________________________________
> > Bioconductor mailing list
> > Bioconductor at stat.math.ethz.ch
> > https://stat.ethz.ch/mailman/listinfo/bioconductor
> 
>