[BioC] Problems in limma package

[Gordon Smyth]

At 05:45 PM 31/03/2004, Binita Dutta wrote:
>2) Please see the attached word file, where the MA plots flips when i 
>normalise between the arrays. I discussed this problem with few users of 
>bioconductor and they were not able to explain this problem.

Before between-array normalization, you have lots of spots in the minimum 
possible intensities, i.e., 1, 2, 3, etc. After between-array 
normalization, you have lots of spots large but equal intensities, hence 
the pattern in your plot.

I don't know why your data specifically has this pattern, but I suspect you 
have a lot of missing values. Between-array normalization cannot be 
recommended in general in the presence of lots of missing values. You need 
to force the intensities to be positive and non-missing.

>3) As you can see below in the table P values are indeed 0.99999 etc  for 
>top differentially expressed genes which is impossible!!!

It is quite possible. Could it be that you are comparing multiple-testing 
adjusted p-values from limma with unadjusted p-values from another program?

>4) As i understand from your explanation i have to
>top<-topTable(fit,number=22680,adjust="fdr")
>
>  and try to subset top and not the function topTable
>
>subset<-subset(top,P.Value<0.01,select=MA$genes)..............am i right?

Sort of. Your 'select=' argument is still wrong. Please read the help for 
'subset'.

Gordon

>Thanks in advance
>
>Binita