[BioC] limma package and related replicate genes problems

[Matthew Ritchie]

Hi Ping,

>
>I am just wondering how to process replicate genes in an array using 
>
>limma package. each gene is printed twice on an array.
>
If the replicate spots are printed in a regular manner on the array, 
then the function lmFit() in limma can be used to summarise the 
log-ratios on each slide before/after normalization.  
For example, if we have non-normalized log-ratios from 6 arrays, with 50 
genes printed in duplicate,

M <- matrix(rnorm(100*6), 100, 6)    # generate random log-ratios

then the following:

design <- diag(1, 6)        # identity design matrix
aveMdups <- lmFit(M, design, ndups=2, spacing=1)
aveM <- aveMdups$coef

will give you averaged log-ratios, stored in aveM.  The identity design 
matrix averages the duplicates for each slide separately, the 'ndups' 
argument gives the number of replicate spots for each gene on the array 
(in your case 2), and 'spacing' specifies how many spots the replicates 
are apart (1 indicates they are side-by-side in a row of a print-tip 
group.  Other alternatives might be the number of spots in a row of a 
print-tip group if the replicate spots are below/above each other in the 
columns of a print-tip group, or half the number of spots on the array 
if they are printed in halves at the top, then the bottom of the array). 
 I hope this helps,

Matt Ritchie
************
PhD Student
Division of Genetics and Bioinformatics
The Walter and Eliza Hall Institute of Medical Research
1G Royal Parade
Parkville, 3050 VIC AUSTRALIA
email: mritchie at wehi.edu.au
web: http://bioinf.wehi.edu.au
************

>For example, the actual number of genes is 19000 in my dataset, since there are replicate genes, I have 38400 spots (genes) in my dataset.
>
>I am just wondering how I can get 19000 genes from these 38400 genes before or after normalization???
>
>It seems that limma has considered this issue, but it integrates with gene selection together (????).  Here I do not want to do that,
>
>I just want to get a dataset which has only 19200 genes (before or after normalization).
>
>Does any one know how to deal with it using limma or other packages???
>
>Thanks,
>
>ping
>