Question: Use of duplicateCorrelation
0
gravatar for Erika Melissari
10.8 years ago by
Erika Melissari240 wrote:
Hello all, I'm studing how to use duplicateCorrelation() of LIMMA package in order to evaluate the between-arrays correlation. I have a simple experiment of direct comparison with dye-swap as follows: targets SlideNumber FileName Cy3 Cy5 Date 1 Ag_1.gpr wt1 1RP 20/9/2008 2 Ag_2.gpr 1RP wt1 20/9/2008 3 Ag_3.gpr wt2 2RP 8/11/2008 4 Ag_4.gpr 2RP wt2 8/11/2008 I use duplicateCorrelation() as follows: design <- c(1,-1,1,-1) > biolrep<-c(1,1,2,2) > corfit<-duplicateCorrelation(MA,design,ndups=1,block=biolrep) > corfit$consensus [1] -0.5543286 The correlation is negative because of dye-swap. Then, I evaluate linear model as explained in limma userguide: > fit<-lmFit(MA,design, block=biolrep,cor=corfit$consensus,weights=NULL) > summary(fit) Length Class Mode coefficients 10807 -none- numeric stdev.unscaled 10807 -none- numeric sigma 10807 -none- numeric df.residual 10807 -none- numeric ndups 1 -none- numeric spacing 1 -none- numeric block 4 -none- numeric correlation 1 -none- numeric cov.coefficients 1 -none- numeric pivot 1 -none- numeric genes 5 data.frame list Amean 10807 -none- numeric method 1 -none- character design 4 -none- numeric fit2<-ebayes(fit) summary(fit2) Length Class Mode df.prior 1 -none- numeric s2.prior 1 -none- numeric s2.post 10807 -none- numeric t 10807 -none- numeric p.value 10807 -none- numeric var.prior 1 -none- numeric lods 10807 -none- numeric > toptable(fit2,adjust="fdr") Error in dim(data) <- dim : attempt to set an attribute on NULL What does it means this error message and, above all, where is the mistake in my analysis procedure? I do not understand why in fit2 there are not any coefficients! I would like to evaluate the dye effect also. How can I do this? I tryed the inclusion of a dye effect coefficient in the design, design<-cbind(Dye=1,design) but when I calculate duplicateCorrelation corfit$consensus is NaN. Is it correct? Thanks very much for any kind of help in advance! Best regards Erika [[alternative HTML version deleted]]
limma • 1.9k views
ADD COMMENTlink modified 10.8 years ago • written 10.8 years ago by Erika Melissari240
Answer: Use of duplicateCorrelation
0
gravatar for Erika Melissari
10.8 years ago by
Erika Melissari240 wrote:
Dear Gordon, I fixed the problem with eBayes....I used ebayes() and not eBayes() and, as reported in LIMMA help, "ebayes is the earlier and leaner function. eBayes is intended to have a more object-orientated flavor as it produces objects containing all the necessary components for downstream analysis. " Then, I fixed the problem by using eBayes() However, I did not manage to fix the problem of duplicateCorrelation() When I define the design including the dye effect and I use this design in duplicateCorrelation(), the calculated consensus correlation is NaN, where without including the dye effect is -0.55 (I have a dye swap design). I have tryed to "unswap" the design also (as you suggested in a previous message), but I have obtained the same result, that is NaN. I have used a different dataset with the same experimental design also and the result was the same: NaN I have tryed to use the simple dye swap design for duplicateCorrelation computation >design<-c(-1,1-1,1) >biolrep<-c(1,1,2,2) >corfit<-duplicateCorrelation(MA,design,ndups=1,block=biolrep) then I have defined a new design >design<-cbind(Dye=1,design) and I have used this together with the correlation previously calculated by duplicateCorrelation (-0.55) for model computing >fit<-lmFit(MA,design, block=biolrep,cor=corfit$consensus,weights=NULL) Is this a right way of mouving aroud the problem? Any suggestion will be appreciated. Erika ----- Original Message ----- From: "Gordon K Smyth" <smyth@wehi.edu.au> To: "Erika Melissari" <erika.melissari at="" bioclinica.unipi.it=""> Cc: "Bioconductor mailing list" <bioconductor at="" stat.math.ethz.ch=""> Sent: Saturday, January 24, 2009 03:40 AM Subject: [BioC] Use of duplicateCorrelation > Dear Erika, > > The output from fit() and eBayes() cannot be as you've given below, > because eBayes() does not remove the $coefficient component of the fitted > model object. Can you please rerun your code from scratch in a new R > session, re-reading the data and so on. > > Best wishes > Gordon > >> Date: Thu, 22 Jan 2009 15:51:43 +0100 >> From: "Erika Melissari" <erika.melissari at="" bioclinica.unipi.it=""> >> Subject: [BioC] Use of duplicateCorrelation >> To: <bioconductor at="" stat.math.ethz.ch=""> >> Message-ID: <00c001c97ca0$f1a27fe0$ba517283 at maanalysis> >> Content-Type: text/plain >> >> Hello all, >> >> I'm studing how to use duplicateCorrelation() of LIMMA package in order >> to evaluate the between-arrays correlation. >> I have a simple experiment of direct comparison with dye-swap as follows: >> targets >> SlideNumber FileName Cy3 Cy5 Date >> 1 Ag_1.gpr wt1 1RP >> 20/9/2008 >> 2 Ag_2.gpr 1RP wt1 >> 20/9/2008 >> 3 Ag_3.gpr wt2 2RP >> 8/11/2008 >> 4 Ag_4.gpr 2RP wt2 >> 8/11/2008 >> >> I use duplicateCorrelation() as follows: >> >> design <- c(1,-1,1,-1) >>> biolrep<-c(1,1,2,2) >>> corfit<-duplicateCorrelation(MA,design,ndups=1,block=biolrep) >>> corfit$consensus >> [1] -0.5543286 >> >> The correlation is negative because of dye-swap. >> Then, I evaluate linear model as explained in limma userguide: >> >>> fit<-lmFit(MA,design, block=biolrep,cor=corfit$consensus,weights=NULL) >>> summary(fit) >> Length Class Mode >> coefficients 10807 -none- numeric >> stdev.unscaled 10807 -none- numeric >> sigma 10807 -none- numeric >> df.residual 10807 -none- numeric >> ndups 1 -none- numeric >> spacing 1 -none- numeric >> block 4 -none- numeric >> correlation 1 -none- numeric >> cov.coefficients 1 -none- numeric >> pivot 1 -none- numeric >> genes 5 data.frame list >> Amean 10807 -none- numeric >> method 1 -none- character >> design 4 -none- numeric >> fit2<-ebayes(fit) >> summary(fit2) >> Length Class Mode >> df.prior 1 -none- numeric >> s2.prior 1 -none- numeric >> s2.post 10807 -none- numeric >> t 10807 -none- numeric >> p.value 10807 -none- numeric >> var.prior 1 -none- numeric >> lods 10807 -none- numeric >>> toptable(fit2,adjust="fdr") >> Error in dim(data) <- dim : attempt to set an attribute on NULL >> >> >> >> What does it means this error message and, above all, where is the >> mistake in my analysis procedure? >> I do not understand why in fit2 there are not any coefficients! >> I would like to evaluate the dye effect also. How can I do this? >> I tryed the inclusion of a dye effect coefficient in the design, >> >> design<-cbind(Dye=1,design) >> >> but when I calculate duplicateCorrelation corfit$consensus is NaN. Is it >> correct? >> >> Thanks very much for any kind of help in advance! >> Best regards >> Erika ---------------------------------------------------------------------- ---------- Database dei virus interno non c aggiornato. Controllato da AVG - http://www.avg.com Versione: 8.0.176 / Database dei virus: 270.9.10/1809 - Data di rilascio: 24/11/2008 09:03 AM
ADD COMMENTlink written 10.8 years ago by Erika Melissari240
Dear Erika, With a dye-effect and only 4 arrays you just don't have enough data to estimate the within-block correlation. There's no right way around not having enough data. I'd be inclined to simplify the model by dropping the blocking. You can't have everything. Best wishes Gordon On Mon, 26 Jan 2009, Erika Melissari wrote: > However, I did not manage to fix the problem of duplicateCorrelation() > > When I define the design including the dye effect and I use this design in > duplicateCorrelation(), the calculated consensus correlation is NaN, where > without including the dye effect is > > -0.55 (I have a dye swap design). > > I have tryed to "unswap" the design also (as you suggested in a previous > message), but I have obtained the same result, that is NaN. > > I have used a different dataset with the same experimental design also and > the result was the same: NaN > > I have tryed to use the simple dye swap design for duplicateCorrelation > computation > >> design<-c(-1,1-1,1) > >> biolrep<-c(1,1,2,2) >> corfit<-duplicateCorrelation(MA,design,ndups=1,block=biolrep) > > then I have defined a new design > >> design<-cbind(Dye=1,design) > > and I have used this together with the correlation previously calculated by > duplicateCorrelation (-0.55) for model computing > >> fit<-lmFit(MA,design, block=biolrep,cor=corfit$consensus,weights=NULL) > > Is this a right way of mouving aroud the problem? > > Any suggestion will be appreciated. > > > > Erika
ADD REPLYlink written 10.8 years ago by Gordon Smyth39k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 242 users visited in the last hour