[BioC] linear model vs. seperate normalization

[Jianping Jin]

Dear list,

I noticed the conversation between John and Wolfgang about how to handle 
human and mouse chips comparison. Wolfgang think that the best way is to 
use linear model that incorporates the species and chiptype effect, etc.

This may also apply to my data, I think. There are 18 chips which are 
derived from two two amplification protocols in the same laboratory for 
some reason. The amplification batch effect is quite obvious in the RLE and 
NUSE plots. I divided files into two groups and re-analyzed. The RLE and 
NUSE plots got much better, especially for the first batch.

I will incorporate the batch effect into limma gle with the data of GCRMA 
normalization for all chips. I also like to try separate GCRMA 
normalization for two batches. Then combine the resulting data and run 
limma without considering batch effect.

The mean intensities after running separate GCRMA are given below:
1. the first group: 17.46
2. the second group: 19.01
3. gcrma using all 18 chips: 18.24

I know the intensity values are pretty low and the difference between group 
1 and 2 is not that large as I expected from RLE plots.

My questions are that
1. do I need to scale the intensities of group 1 and 2 to a same level?
2. if so, is there any program in BioConductor that I can use to do that?

Thanks!

Jianping

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
x Jianping Jin Ph.D.            x
x Bioinformatics scientist      x
x Center for bioinformatics     x
x 3133 Bioinformatics Building  x
x CB# 7104                      x
x University of North Carolina  x
x Chapel Hill, NC 27599         x
x Tel: (919)843-6105            x
x Fax: (919)843-3103            x
x E-mail: jjin at email.unc.edu    x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx