[BioC] Data analysis

[Gordon Smyth]

At 02:20 AM 14/10/2003, Jason Skelton wrote:
>I have a question about data analysis after normalisation
>I have normalised in limma and applied the generalized least squares 
>linear models to my data with very nice results!
>
>My experiment is a Three sample experiment
>(Three different treatments)compared to a commmon reference.
>The Three samples have six slides per experiment, 3 in one dye orientation 
>and 3 dye swapped to give 18 slides in total.
>I have approx 3500 genes in duplicate on my array at present.
>
>Currently I have normalised all three sets of data seperately but would 
>like to be able to compare the three data sets.
>I was thinking of using the mva functions like dist/hclust etc.
>
>My questions:
>Is this the best way of comparing this data or are there other/better 
>methods that could be used that anyone has had experience with.

I would use the limma commands lmFit (or lm.series or gls.series) followed 
by makeContrasts, eBayes and classifyTests. See the earliers posts:

   https://stat.ethz.ch/pipermail/bioconductor/2003-September/002406.html
   https://stat.ethz.ch/pipermail/bioconductor/2003-September/002405.html

This would allow you to answer the following questions:

1. Which genes are differentially expressed between each pair of treatments?
2. Which genes are differentially expressed between each treatment and the 
reference?
3. Which genes show _any_ differences between the treatments?

Your problem doesn't sound like a cluster analysis problem to me.

Cheers
Gordon

>e.g. similar to the two-sample experiment example in limma user guide
>where results from the  linear model and ebayes are displayed with a heatmap ?
>(sorry I'm presuming that this is the kind of thing I should be doing ?)
>
>Also I'm presuming the data I want to use for these methods are the 
>normalised $M values ? OR do I want to use the results from 
>gls.series/lm.series and ebayes for a 3 sample comparison ?
>
>please could someone give me an example of the best method they recommend 
>with some commands that I could try using...
>
>Thanks very much to anyone who can help
>
>Jason