Hi Yanju,
Two suggestions - 1) The code I gave you before was written as if your
reference was in the green (Cy3) channel. However, based on the
results of
your 'modelMatrix(targets, ref="gDNA")' command, your reference is in
the
red (Cy5) channel. Therefore, you would have to reverse some of the
commands where appropriate (e.g., use method "Rquantile" and replace M
values with G values).
2) Find a local statistician to consult about the analysis, because it
appears you have a 2 x 7 factorial design (2 strains and 7 timepoints
= 14
treatment groups total). There are variety of ways to analyze this
experimental design, depending on what all you want to know from the
data.
If you really only wanted to know which genes were different between
mu &
wt at each time point independently, then you could analyze the
arrays
from each time point separately. However, there is so much more
information
to be gained from this data set, which is why I suggest you consult a
local
statistician.
Best of luck,
Jenny
At 10:54 AM 11/21/2006, yanju wrote:
>Hello Jenny,
>
>I adapted my code according to your suggestion. Then at some time
points,
>the results showed that the most differently expressed genes are
markers.
>This is every werld.
>
>And It doesnt matter if I change -1 in the design matrix to 1 (my
method:
>new design matrix=old design matrix* -1, old design matrix was
derived
>from modelMatrix function) or not. I mean this didnt effect my
results.
>
>Since I could not figure it out, I paste my code here. Hope you could
tell
>me what's wrong with my program. Basic information of the data:
>Two-channle array, 7 time points from 16-72h, at each time point
there are
>some repelicants. Aim is to detect the different expressed genes at
each
>time points.
>
> From the very begin of the code:
>
>targets<-readTargets("target_new_16_72.txt")
>rg<-read.maimages(targets, source="genepix",wt.fun=wtflags(0.1))
> # read targets and genepix files
>
>rgc<-backgroundCorrect(rg, method="half")
> # bacground correction
>
>MA.Gquant<-normalizeBetweenArrays(rgc, method="Gquantile")
>RG.Gquant<-RG.MA(MA.Gquant)
>MA.fake<-MA.Gquant
>MA.fake$M<-log2(RG.Gquant$R)
> #normalization
>
>design<-modelMatrix(targets, ref="gDNA")
>design_revise<-design*-1
> #design was similar like follows.
> # wt16 wt20 wt24
> #[1,] -1 0 0
> #[2,] 0 -1 0
> #[3,] 0 0 -1
> #Then it was multiply by -1 to have the positive value.
>
>
> fit<-lmFit(MA.fake,design_revise)
> cont.matrix<-makeContrasts(MUvsWT16=mu16-wt16,
> MUvsWT20=mu20-wt20,
> MUvsWT24=mu24-wt24, MUvsWT36=mu36-wt36,
> MUvsWT48=mu48-wt48,
> MUvsWT60=mu60-wt60, MUvsWT72=mu72-wt72,levels=design_revise)
> fit2<-contrasts.fit(fit,cont.matrix)
> fit2<-eBayes(fit2)
> #fit the data to linear model and Bayes statistical summary
>
>result20<-topTable(fit2,coef=2, number=20,adjust="BH")
> #detecting the top20 differently expressed genes at time
point20.
> #But as I said, most of the top20 were markers or the "no
spot"
>
>Hope you could help me figure out the problems. I really appreciate
your
>help. Thanks.
>
>Regards,
>Yanju
>
>
Jenny Drnevich, Ph.D.
Functional Genomics Bioinformatics Specialist
W.M. Keck Center for Comparative and Functional Genomics
Roy J. Carver Biotechnology Center
University of Illinois, Urbana-Champaign
330 ERML
1201 W. Gregory Dr.
Urbana, IL 61801
USA
ph: 217-244-7355
fax: 217-265-5066
e-mail: drnevich at uiuc.edu