[BioC] 2x2 factorial loop without common reference (pool)
Naomi Altman
naomi at stat.psu.edu
Sun Apr 23 23:03:13 CEST 2006
A single channel analysis does not treat a
2-channel array like 2 single-channel arrays. It treats the array as a block.
The single channel analysis is more powerful,
unless the errors are perfectly correlated within
spot. Also, it is much easier to handle
complicated designs, where complicated means
pretty much everything that is not a paired t-test or reference design.
Using the difference is channels is certainly
valid. I cannot see why people want to throw
away the information about the treatment means,
however, which is often important to the biologists I collaborate with.
--Naomi
At 04:52 PM 4/23/2006, francois fauteux wrote:
>Hi again;
>
>Trying to work this out in the "two channels" way (I don't get clearly
>what are the benefits of treating a two-color array as if it was two
>one-color arrays)... If really necessary and clearly justified, I
>might whant to switch to this type of analysis, but otherwise would
>like to stay in what's more 'standard'.
>
>Other try, but new design get error messages when lmfit, coefficients
>not estimable 1 out of 2:
>
> > fit4 <- lmFit(MAq, weights=w, design4)
>Coefficients not estimable: b.a c.a d.c
>
> > design4
> a.b b.a a.c c.a b.d d.b c.d d.c
> [1,] 1 -1 0 0 0 0 0 0
> [2,] 1 -1 0 0 0 0 0 0
> [3,] 1 -1 0 0 0 0 0 0
> [4,] -1 1 0 0 0 0 0 0
> [5,] -1 1 0 0 0 0 0 0
> [6,] -1 1 0 0 0 0 0 0
> [7,] 0 0 1 -1 0 0 0 0
> [8,] 0 0 1 -1 0 0 0 0
> [9,] 0 0 1 -1 0 0 0 0
>[10,] 0 0 -1 1 0 0 0 0
>[11,] 0 0 -1 1 0 0 0 0
>[12,] 0 0 -1 1 0 0 0 0
>[13,] 0 0 0 0 0 -1 0 0
>[14,] 0 0 0 0 1 -1 0 0
>[15,] 0 0 0 0 1 -1 0 0
>[16,] 0 0 0 0 1 1 0 0
>[17,] 0 0 0 0 -1 1 0 0
>[18,] 0 0 0 0 -1 1 0 0
>[19,] 0 0 0 0 -1 0 1 -1
>[20,] 0 0 0 0 0 0 1 -1
>[21,] 0 0 0 0 0 0 1 -1
>[22,] 0 0 0 0 0 0 -1 1
>[23,] 0 0 0 0 0 0 -1 1
>[24,] 0 0 0 0 0 0 -1 1
>
>Why are Coefficients not estimable??? Thought this design would take
>into account the dye effects...
>
>Second design try:
>
>design3 <- modelMatrix(targets,ref="a")
> > contrast.matrix <- makeContrasts(b,c,d-b,d-c,
>+ levels=design3)
> > contrast.matrix
> b c d - b d - c
>c 0 1 0 -1
>b 1 0 -1 0
>d 0 0 1 1
>
>This seems correct a priori. If you could confirm that it is OK, and
>that this is indeed the best way of working things out...
>
>Many thanks, regards.
>
>François
>
>On 4/23/06, Naomi Altman <naomi at stat.psu.edu> wrote:
> > I would use single channel analysis for
> > this. The only problem is that Limma allows only
> > 1 level of random effects. Hence, you will need to average the dye-swaps.
> >
> > Anyways with single channel analysis, you have a
> > balanced incomplete block design with factorial
> > treatments, and the analysis is much simplified.
> >
> > --Naomi
> >
> > At 01:41 PM 4/23/2006, francois fauteux wrote:
> > >Hi;
> > >
> > >We are doing an experiment with agilent 44K (3 biological reps,
> > >complete dye-swap):
> > >
> > >a - control
> > >b - treatment 1
> > >c - treatment 2
> > >d - treatment 1 + treatment 2
> > >
> > >and I would like to output evidence of the interaction between two
> > >treatments and effect on gene expression.
> > >
> > >24 chips:
> > >
> > >SlideNumber Cy3 Cy5
> > >1 a1 b1
> > >2 a2 b2
> > >3 a3 b3
> > >4 b1 a1
> > >5 b2 a2
> > >6 b3 a3
> > >7 a1 c1
> > >8 a2 c2
> > >9 a3 c3
> > >10 c1 a1
> > >11 c2 a2
> > >12 c3 a3
> > >13 b1 d1
> > >14 b2 d2
> > >15 b3 d3
> > >16 d1 b1
> > >17 d2 b2
> > >18 d3 b3
> > >19 c1 d1
> > >20 c2 d2
> > >21 c3 d3
> > >22 d1 c1
> > >23 d2 c2
> > >24 d3 c3
> > >
> > >I've done several tests with limma to isolate significant results in
> > >the following:
> > >1- a vs b;
> > >2- a vs c;
> > >3- b bs d;
> > >4- c vs d;
> > >
> > >with this "targets.txt":
> > >
> > >SlideNumber Cy3 Cy5
> > >1 a b
> > >2 a b
> > >3 a b
> > >4 b a
> > >5 b a
> > >6 b a
> > >7 a c
> > >8 a c
> > >9 a c
> > >10 c a
> > >11 c a
> > >12 c a
> > >13 b d
> > >14 b d
> > >15 b d
> > >16 d b
> > >17 d b
> > >18 d b
> > >19 c d
> > >20 c d
> > >21 c d
> > >22 d c
> > >23 d c
> > >24 d c
> > >
> > >First option:
> > >
> > > > f <- paste(targets$Cy3, targets$Cy5, sep = ".")
> > > > f <- factor(f, levels = c("a.b", "b.a",
> > > "a.c", "c.a", "b.d", "d.a", "c.d", "d.a"))
> > > > design1 <- model.matrix(~0 + f)
> > >
> > > > design
> > > a.b b.a a.c c.a b.d d.b c.d d.c
> > >1 1 0 0 0 0 0 0 0
> > >2 1 0 0 0 0 0 0 0
> > >3 1 0 0 0 0 0 0 0
> > >4 0 1 0 0 0 0 0 0
> > >5 0 1 0 0 0 0 0 0
> > >6 0 1 0 0 0 0 0 0
> > >7 0 0 1 0 0 0 0 0
> > >8 0 0 1 0 0 0 0 0
> > >9 0 0 1 0 0 0 0 0
> > >10 0 0 0 1 0 0 0 0
> > >11 0 0 0 1 0 0 0 0
> > >12 0 0 0 1 0 0 0 0
> > >13 0 0 0 0 1 0 0 0
> > >14 0 0 0 0 1 0 0 0
> > >15 0 0 0 0 1 0 0 0
> > >16 0 0 0 0 0 1 0 0
> > >17 0 0 0 0 0 1 0 0
> > >18 0 0 0 0 0 1 0 0
> > >19 0 0 0 0 0 0 1 0
> > >20 0 0 0 0 0 0 1 0
> > >21 0 0 0 0 0 0 1 0
> > >22 0 0 0 0 0 0 0 1
> > >23 0 0 0 0 0 0 0 1
> > >24 0 0 0 0 0 0 0 1
> > >
> > >This gives significant results for each one of the "levels" but does
> > >not take into account the dye-swap (i.e "a.b" and "b.a" are considered
> > >independent).
> > >
> > >Other tested option is:
> > > > design2 <- modelMatrix(targets,ref="a")
> > >
> > > > design
> > > p s sp
> > >ab1 0 1 0
> > >ab2 0 1 0
> > >ab3 0 1 0
> > >ba1 0 -1 0
> > >ba2 0 -1 0
> > >ba3 0 -1 0
> > >ac1 1 0 0
> > >ac2 1 0 0
> > >ac3 1 0 0
> > >ca1 -1 0 0
> > >ca2 -1 0 0
> > >ca3 -1 0 0
> > >bd1 0 -1 1
> > >bd2 0 -1 1
> > >bd3 0 -1 1
> > >db1 0 1 -1
> > >db2 0 1 -1
> > >db3 0 1 -1
> > >cd1 -1 0 1
> > >cd2 -1 0 1
> > >cd3 -1 0 1
> > >dc1 1 0 -1
> > >dc2 1 0 -1
> > >dc3 1 0 -1
> > >
> > >This gives results for "b" effect, "c" effect, and "d" effect.
> > >However, I could'nt get results for the 4 comparisons of interest
> > >(even though the matrix is coherent).
> > >
> > >Questions:
> > >
> > >1 - What would be the best option (design and operations) to get to
> > >contrasts of interest considering that the experiment has a 4
> > >treatments in a factorial design without common reference (a vs b, a
> > >vs c, b vs d, c vs d) and taking into account the dye-effect;
> > >
> > >2- Is this method (4 contrasts) the best one considering that
> > >treatment "d" is a combination of treatments "b" and "c" (factorial
> > >type design). How could one directly get to identify genes
> > >differentially expressed due to the interaction between treatment "b"
> > >and treatment "c" (i.e effect of "d" over "b" and "c").
> > >
> > >In Limma Users Guide and elsewhere on this forum, I could not find a
> > >clear description of how this type of analysis should be performed,
> > >even though it is a simple design (i.e 2X2 factorial without a common
> > >reference - two color arrays - complete dye swap).
> > >
> > >Thanks for your time, best regards.
> > >
> > >François Fauteux
> > >Étudiant à la maîtrise en biologie végétale
> > >Centre de recherche en horticulture
> > >Université Laval
> > >francois.fauteux at gmail.com
> > >
> > >_______________________________________________
> > >Bioconductor mailing list
> > >Bioconductor at stat.math.ethz.ch
> > >https://stat.ethz.ch/mailman/listinfo/bioconductor
> > >Search the archives:
> > >http://news.gmane.org/gmane.science.biology.informatics.conductor
> >
> > Naomi S. Altman 814-865-3791 (voice)
> > Associate Professor
> > Dept. of Statistics 814-863-7114 (fax)
> > Penn State University 814-865-1348 (Statistics)
> > University Park, PA 16802-2111
> >
> >
Naomi S. Altman 814-865-3791 (voice)
Associate Professor
Dept. of Statistics 814-863-7114 (fax)
Penn State University 814-865-1348 (Statistics)
University Park, PA 16802-2111
More information about the Bioconductor
mailing list