[BioC] Limma double paired analysis

[john seers (IFR)]

Hi James

Yes, that makes sense now.

Thanks very much. I will give it a try now.

Regards


John Seers


 
---

John Seers
Bioinformatics & Statistics
Institute of Food Research
Norwich Research Park
Colney
Norwich
NR4 7UA
 
Location: IFR1 N102
PC Machine ID: N198


tel +44 (0)1603 251497
fax +44 (0)1603 507723
e-mail john.seers at bbsrc.ac.uk                         
e-disclaimer at http://www.ifr.ac.uk/edisclaimer/ 
 
Web sites:

www.ifr.ac.uk   
www.foodandhealthnetwork.com

-----Original Message-----
From: James W. MacDonald [mailto:jmacdon at med.umich.edu] 
Sent: 19 September 2008 13:52
To: john seers (IFR)
Cc: bioconductor at stat.math.ethz.ch
Subject: Re: [BioC] Limma double paired analysis

Hi John,

john seers (IFR) wrote:
> 
> Hello All
> 
> I am having some difficulty in trying to use limma for a paired 
> analysis. Can anybody suggest an approach that would work?
> 
> The arrays are Affymetrix.
> 
> The experiment arrays look like:
> 
> Volunteer1              Ctrl <---------->Placebo<-------Time
> passes------>Ctrl<------------>Drug
> Volunteer2              Ctrl <---------->Placebo<-------Time
> passes------>Ctrl<------------>Drug
> Volunteer3              Ctrl <---------->Drug<----------Time
> passes------>Ctrl<------------>Placebo
> ...
> 
> That is a control sample is taken. A treatment is given - Placebo or 
> Drug. Another sample is taken. A suitable time period passes and this 
> is repeated but with the Placebo or Drug treatments reversed.
> 
> How can this be analysed?
> 
> Looking at the Limma User Guide 8.3 Paired Samples looks to be a good 
> start. But is it possible to do some form of double pairing analysis?
> That is I can pair the Ctrl arrays with their paired treatment. But 
> then I want to pair the Placebo with the Drug. How can I do this using
limma?

What you want to do is identical to the part of the User's Guide that
you quote. The only difference is that the blocking will be of size four
instead of size two. So the 'block' factor would contain four 1's
corresponding to volunteer 1, then four 2's for volunteer 2, etc.

What this does is to remove the average expression value for each
volunteer so you can compare the different treatments directly. The
assumption here being that the only difference between the patients is
their average expression level (e.g., the treatment differences are
comparable, but the baseline expression levels may be different).

You can do a slightly more sophisticated analysis by using
duplicateCorrelation() and the block argument of lmFit()to fit a glm
rather than a fixed effects model as well.

Best,

Jim


> 
> 
> ======================================================================
> ==
> =============
> 
> So far I have something like the following, if it helps. How can I 
> factor in that Drug Treatment and the Placebo treatment are paired?
> 
> 
> Treatment<-c("Drug", "Placebo", "ACtrl", "ACtrl", "ACtrl", "Drug", 
> "ACtrl",
>     "Placebo", "ACtrl", "Placebo", "ACtrl", "Drug", "Placebo", 
> "ACtrl", "ACtrl",
>     "Drug", "Drug", "ACtrl", "ACtrl", "Placebo", "Placebo", "ACtrl",

>     "Drug", "ACtrl", "Drug", "Placebo", "ACtrl", "ACtrl", "ACtrl", 
> "Placebo",
>     "ACtrl", "Drug", "ACtrl", "Placebo", "Drug", "ACtrl", "Placebo", 
> "ACtrl",
>     "ACtrl", "Drug", "Placebo", "ACtrl", "Drug", "ACtrl", "ACtrl",
> "ACtrl",    
>     "Drug", "Placebo")
> 
> 
> Pairing<-c("A13", "B13", "B13", "A13", "A17", "B17", "B17", "A17", 
> "B8", "A8",  "A8",
>     "B8",  "B18", "B18", "A18", "A18", "A6",  "B6",  "A6",  "B6", 
> "A33", "B33",
>     "B33", "A33", "A27", "B27", "A27", "B27", "A22", "B22", "B22", 
> "A22", "B11", "A11",
>     "B11", "A11", "A28", "A28", "B28", "B28", "A16", "A16", "B16", 
> "B16", "A20", "B20", "A20", "B20")
>    
> 
> Pairing<-factor(Pairing)
> Treatment<-factor(Treatment)
> design<-model.matrix(~ 0 + Pairing + Treatment) fit<-lmFit(eset, 
> design) contrast.matrix<-makeContrasts(AbovePlacebo = TreatmentDrug - 
> TreatmentPlacebo, levels=design) fit2<-contrasts.fit(fit, 
> contrast.matrix)
> eb<-eBayes(fit2)
> tt<-topTable(eb)
> 
> Any help gratefully received.
> 
> 
> Regards and thanks
> 
> 
> John Seers
> 
> 
> 
> 
> ---
> 
> John Seers
> Bioinformatics & Statistics
> Institute of Food Research
> Norwich Research Park
> Colney
> Norwich
> NR4 7UA
> 
> Location: IFR1 N102
> PC Machine ID: N198
> 
> 
> tel +44 (0)1603 251497
> fax +44 (0)1603 507723
> e-mail john.seers at bbsrc.ac.uk                        
> e-disclaimer at http://www.ifr.ac.uk/edisclaimer/
> 
> Web sites:
> 
> www.ifr.ac.uk
> www.foodandhealthnetwork.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

--
James W. MacDonald, M.S.
Biostatistician
Hildebrandt Lab
8220D MSRB III
1150 W. Medical Center Drive
Ann Arbor MI 48109-0646
734-936-8662