[BioC] Re: marrayClasses algorithm

[Jean Yee Hwa Yang]

Hi Joyce,

> 1, I use read.marrayRaw() to read into my data into R, and found I have a lot 
> of NA value, I would like to know what is the algorithm to calcualte this 
> value. It is log ratio, do you also have certain kind of thresh value to cut 
> off?

read.marrayRaw doesn't do any thresholding or cutoff.  The M values are
calculated by
M = log(Rf - Rb / Gf- Gb)
Now if your Rb > Rf than is will generate NA values.


> 2, I use loess algorithm to normalize my data, and did correlation between 
> replicates, they all have pretty good correlation, but I found that among 
> different samples(not replicates), their correlation are still good, which 
> does not make sense. 

Are you using a common reference design?  Are you calculating correlation
between log-ratios or log-intensities?  You will see high correlation with
log-intensities.  But for log-ratios, it will depends on  the number of
your DE genes.

> 3, I use maPlot() to plot pre-normalization and post-normalization(which are 
> implemented in PrintTip algorithm). I did see some changes between, but I saw 
> the line yield down, but dot just yield a little bit more.

I don't quite follow your question.  The loess line should look horizontal
post-normalization.

Will take a look at your data files.

Cheers

Jean