[BioC] Still confused on how to identify differentially expressed genes using limma

[Stanley Ng]

Hi, 

Please bear with me. The design model (common reference and dye-swap) is as
follows:

Array 1: Wild type cells at 0h (Cy5) vs Pooled RNA (Cy3) i.e. WT.0h.Cy5 vs
PR.Cy3
Array 2: Pooled RNA (Cy5) vs Wild type cells at 0h (Cy3) i.e. PR.Cy5 vs
WT.0h.Cy3 
Array 3: Wild type cells transfected with empty plasmid at 8h (Cy5) vs
Pooled RNA (Cy3) i.e. WT.EP.8h.Cy5 vs PR.Cy3
Array 4: Pooled RNA (Cy5) vs Wild type cells transfected with empty plasmid
at 8h (Cy3) i.e. PR.Cy5 vs WT.EP.8h.Cy3 
Array 5: Wild type cells transfected with plasmid expressing P53 at 8h (Cy5)
vs Pooled RNA (Cy3) i.e. WT.EP+P53.8h.Cy5 vs PR.Cy3
Array 6: Pooled RNA (Cy5) vs Wild type cells transfected with plasmid
expressing P53 at 8h (Cy3) i.e. PR.Cy5 vs WT.EP+P53.8h.Cy3  
Array 7: Wild type cells transfected with empty plasmid at 16h (Cy5) vs
Pooled RNA (Cy3) i.e. WT.EP.16h.Cy5 vs PR.Cy3
Array 8: Pooled RNA (Cy5) vs Wild type cells transfected with empty plasmid
at 16h (Cy3) i.e. PR.Cy5 vs WT.EP.16h.Cy3 
Array 9: Wild type cells transfected with plasmid expressing P53 at 16h
(Cy5) vs Pooled RNA (Cy3) i.e. WT.EP+P53.16h.Cy5 vs PR.Cy3
Array 10: Pooled RNA (Cy5) vs Wild type cells transfected with plasmid
expressing P53 at 16h (Cy3) i.e. PR.Cy5 vs WT.EP+P53.16h.Cy3 

I need to identify two sets of differentially expressed genes induced solely
by the over-expression of P53 at 8h, and at 16h. I took the following steps
using the limma package:

1) Background correct each array
2) Normalize each array
3) Normalize array 1,3,5,7,9 (array 2,4,6,8,10) to have identical Cy3 (Cy5)
distribution; Pooled RNA serves as baseline  
4) To eliminate the effect of empty plasmid at 8h and 16, WT.P53.8h.Cy5 =
WT.EP+P53.8h.Cy5 - WT.EP.8h.Cy5 + WT.0h.Cy5; similarly for Cy3 at 8h and 16h
5) Create the following RGlist:
	WT.P53.8h.Cy5 vs WT.0h.Cy3
	WT.0h.Cy5 vs WT.P53.8h.Cy3 	 
	WT.P53.16h.Cy5 vs WT.0h.Cy3
	WT.0h.Cy5 vs WT.P53.16h.Cy3 	 
6) Create the design matrix given WT.0h.Cy5 and WT.0h.Cy3 as the reference
7) Create contrast matrix for WT.P53.8h - WT.0h and WT.P53.16h - WT.0h 
8) Use lmfit, and ebayes.

Couple of questions:
A) Are my steps reasonably sound ?
B) Should log2 transformed be used for step 4 ? In my perspective, I think
not to use the log2 transformed values, as the difference in intensities is
to be computed. Please comment 
C) Better alternative you may have to what I have ? 

Thanks
Stanley