[BioC] Time course experiment with limma

[Naomi Altman]

Dear Cecilia
You cannot analyze this design properly in limma because you have 3 
sources of correlation - 2 samples from the same RNA, 2 channels on 
the same  array, and multiple time points on the same plant.  Limma 
allows only 1 source of correlation.

I have not used MAANOVA, but I am familiar with the statistical 
method used, and to me it seems like a very good alternative to 
limma.  (NOT MANOVA, which is different.)

2 notes based on my experience:  1) You are wasting half your arrays 
by having 2 samples from each RNA.
             2) If you are taking multiple samples from the same 
plants, you need to be sure that the wound response has died out 
between time points.

--Naomi


At 03:40 PM 11/28/2006, Cecilia McGregor wrote:
>I'm planning a time course experiment and was told that with the 
>design I plan I would not be able to use limma for analysis and need 
>to use MANOVA. I want to know whether this is true and whether there 
>is a different experimental design I can use that would make it 
>possible for me to use limma. (Simply 'cause I've used limma before 
>, but never used MANOVA). Any other comments about the experiment 
>and experimental design is also welcome.
>
>Here follows a somewhat lenthy description of the experiment.
>
>The  treatments are:
>(1) uninfected plants
>(2) plants infected with SPFMV-RC (a strain that leads to SPVD in 
>dual infections) alone,
>(3) SPFMV-C (a strain that does not cause SPVD in dual infections) alone,
>(4) plants infected with SPCSV alone,
>(5) plants infected with SPFMV-RC and SPCSV together (SPVD),
>(6) plants infected with SPFMV-C and SPCSV together (No SPVD).
>
>We are trying to figure out what it is that happens to the plants 
>defense system that allows for the severe disease in the dual 
>infection (SPFMV-RC and SPCSV).
>
>But we are also interested in the development of the disease over time.
>
>We therefore interested in both comparisons of treatment groups 
>within each time point and comparison of time points in each group.
>
>The 5 time points will be: 2 days after infections (DAI), 5 DAI, 10 
>DAI, 15 DAI, 20 DAI).
>
>We are collecting from the same individuals (plants) for all 
>timepoints. So if I say we have 3 biological replications per 
>treatment, it means that on Day 0, I inoculate 3 plants for each 
>treatment, and these 3 plants are used for all samples throughout 
>the time of the experiment. But when I sample I take whole leaves, 
>so that means that I have to take different leaves every time I 
>samples. So, same plants for all timepoints, but different leaves.
>
>In the greenhouse the plants are in a completely randomized design.
>
>Recap of information I gave in previous e-mail
> > - I have 6 treatments (including the untreated),
> > - 3 biological replicates per treatment per timepoint
> > - 5 timepoints
> > - The problem is that we have only 60 arrays! The original 
> experiment was planned with 120 arrays, but the price from the 
> supplier doubled from a year ago when we did our other experiments.
> > - These are two color cDNA arrays.
>
>
>The experimental design that I plan to use:
>(1) Loops comparing DAI within each treatment: For each treatment, 
>choose a single biological replication and connect its mRNA samples 
>from different DAI using a loop. Note that each mRNA sample goes in 
>two slides (with alternating labeling). Each loop will use 5 slides, 
>and you'll have 6 of those loops with a total of 30 slides comparing DAIs.
>(2) Loops comparing treatments within each time point: For each time 
>point (DAI), choose two biological replications from each treatment 
>and connect the mRNA samples following the structure given below. 
>This is also a loop, but not a "connected loop" as above. This kind 
>of loop favors biological replication over technical replication. 
>Here, note that each mRNA sample goes in one slide only, but there 
>is still balance in labeling across treatments. There will be 5 
>loops of 6 slides each, with a total of 30 slides for comparing treatments.
>
>Can this type of design be analyzed with limma? I was told that I 
>would have to use MANOVA. Is there a more appropriate design for 
>this experiment, or a different  appropriate design that could be 
>analyzed with limma?
>
>Any help would be very much appreciated.
>
>Cecilia McGregor
>
>Post-Doc
>Sweetpotato Breeding and Genetics Lab
>JC Miller Hall room 236
>Louisiana State University
>Baton Rouge
>LA, 70803
>USA
>
>Phone: (225) 578 2173
>
>_______________________________________________
>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