[BioC] Testing two hypothesis by linear models

[Ron Ophir]

Dear limma users,
In our study we would like to understand which gene differentially
expressed in which arabidopsis tissue. We extract the RNA from leaves
from various developmental stages in way we cannot completely separate
the tissues. However each RNA extraction should include transcripts from
different mix of tissues. I think that Linear Models is perfect for such
research since we can decide the expression measurement into its tissue
components based on a model. The problem is that we are not completely
sure as for the existence of one of tissues in one of the RNA extracts.
Here is an example of the design
	LP	ML	ADL	ABL	T	S	B
YE1ATH1.CEL	1	1	1	1	1	1	1
YE10ATH1.CEL	1	1	1	1	1	1	1
YE14ATH1.CEL	0	1	1	1	1	0	0
YE15ATH1.CEL	1	0	1	1	1	1	1
YE16ATH1.CEL	0	1	1	1	1	0	0
YE17ATH1.CEL	1	0	1	1	1	1	1
YE2ATH1.CEL	1	0	1	0	1	0	0
YE20ATH1.CEL	1	0	1	1	1	1	1
YE21ATH1.CEL	1	0	1	1	0	1	1
YE22ATH1.CEL	1	0	1	1	1	0	1
YE27ATH1.CEL	1	0	1	1	1	0	1
YE3ATH1.CEL	1	0	0	1	0	1	0
YE5ATH1.CEL	1	0	0	1	0	1	0
YE6ATH1.CEL	1	0	1	0	1	0	0
YE7ATH1.CEL	1	1	1	1	1	1	1
YE9ATH1.CEL	1	0	1	1	1	1	1

where the parameter (column headers) are the tissues. We have
replicates here for example YE1ATH1.CEL is duplicate of YE10ATH1.CEL so
maybe it was more correct to set the matrix values there into 0.5 and
0.5 for each.
The question is if we change one of the tissue design what would be the
criteria to design which model is better. Is it something that can be
deduced from the coefficients after fitting the models or it is
something we should deduce from the expected expression of certain
genes?
Thanks,
Ron