Dear Jose, The warning from lmFit was introduced a while ago (a year or so) as a result of a user request on this mailing list. The warning is to remind people that, when some of the coefs for a particular probe become NA but others are estimable, the interpretation of the remaining coefs is changed for that probe. This is an unavoidable characteristic of missing values. I agree the warning doesn't make sense when the coefs are either all NA or all estimable, as in your example below. I'll fix that. Best wishes Gordon > Date: Wed, 04 Mar 2009 11:06:42 +0000 > From: J.delasHeras at ed.ac.uk > Subject: [BioC] LIMMA, behaviour of 'lmFit': warnings > To: Bioconductor mailing list <bioconductor at="" stat.math.ethz.ch=""> > > Hi, > > I am using the Limma package and I am getting warnings when trying to > fit a linear model to my data in an "MAList", using a vector of > weights (0 or 1) to discount genes that do not pass the low intensity > filter in *all* my arrays. > > I made a little example with made up data here that reproduces teh warnings: > > # Make an MAList with 3 columns of 100 genes: > MA3<-new("MAList") > MA3$M<-matrix(rnorm(300), ncol=3) > MA3$A<-matrix(abs(rnorm(300)), ncol=3) > > # create a weights vector, all 1 except for three elements turned to 0. > low3<-rep(1, times=100) > low3[c(10, 50, 90)]<-0 > > # no dye swap, straight replicates > design<-c(1,1,1) > > # running the following command (no weights) works ok: > fit<-lmFit(MA3, design=design, weights=NULL) # ok > > > # if I try to use the weights, I get a warning: > fit<-lmFit(MA3, design=design, weights=low3) # warnings > > Warning message: > In lmFit(MA3, design = design, weights = low3) : > Some coefficients not estimable: coefficient interpretation may vary. > > but the coefficients for the zero weighted elements are NA, however. > > What is the reason for the warning? Shall I be concerned? > > In my actual data about 25% of the genes are weighted zero. > > This is my sessionInfo(): > >> sessionInfo() > R version 2.7.0 (2008-04-22) > i386-pc-mingw32 > > locale: > LC_COLLATE=English_United Kingdom.1252;LC_CTYPE=English_United > Kingdom.1252;LC_MONETARY=English_United > Kingdom.1252;LC_NUMERIC=C;LC_TIME=English_United Kingdom.1252 > > attached base packages: > [1] stats graphics grDevices datasets tcltk utils methods base > > other attached packages: > [1] limma_2.14.0 svSocket_0.9-5 svIO_0.9-5 R2HTML_1.59 > svMisc_0.9-5 svIDE_0.9-5 > > loaded via a namespace (and not attached): > [1] tools_2.7.0 >> > > > Jose > > -- > Dr. Jose I. de las Heras Email: J.delasHeras at ed.ac.uk > The Wellcome Trust Centre for Cell Biology Phone: +44 (0)131 6513374 > Institute for Cell & Molecular Biology Fax: +44 (0)131 6507360 > Swann Building, Mayfield Road > University of Edinburgh > Edinburgh EH9 3JR > UK > > -- > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336.

Dear Gordon, thanks for that, now I see what the warning was about. I was worried I may have misunderstood something. As you say, it probably is best to change the warning when all coefs are NA for a given gene. regards, Jose Quoting Gordon K Smyth <smyth at="" wehi.edu.au="">: > Dear Jose, > > The warning from lmFit was introduced a while ago (a year or so) as a > result of a user request on this mailing list. > > The warning is to remind people that, when some of the coefs for a > particular probe become NA but others are estimable, the interpretation > of the remaining coefs is changed for that probe. This is an > unavoidable characteristic of missing values. > > I agree the warning doesn't make sense when the coefs are either all NA > or all estimable, as in your example below. I'll fix that. > > Best wishes > Gordon > >> Date: Wed, 04 Mar 2009 11:06:42 +0000 >> From: J.delasHeras at ed.ac.uk >> Subject: [BioC] LIMMA, behaviour of 'lmFit': warnings >> To: Bioconductor mailing list <bioconductor at="" stat.math.ethz.ch=""> >> >> Hi, >> >> I am using the Limma package and I am getting warnings when trying to >> fit a linear model to my data in an "MAList", using a vector of >> weights (0 or 1) to discount genes that do not pass the low intensity >> filter in *all* my arrays. >> >> I made a little example with made up data here that reproduces teh warnings: >> >> # Make an MAList with 3 columns of 100 genes: >> MA3<-new("MAList") >> MA3$M<-matrix(rnorm(300), ncol=3) >> MA3$A<-matrix(abs(rnorm(300)), ncol=3) >> >> # create a weights vector, all 1 except for three elements turned to 0. >> low3<-rep(1, times=100) >> low3[c(10, 50, 90)]<-0 >> >> # no dye swap, straight replicates >> design<-c(1,1,1) >> >> # running the following command (no weights) works ok: >> fit<-lmFit(MA3, design=design, weights=NULL) # ok >> >> >> # if I try to use the weights, I get a warning: >> fit<-lmFit(MA3, design=design, weights=low3) # warnings >> >> Warning message: >> In lmFit(MA3, design = design, weights = low3) : >> Some coefficients not estimable: coefficient interpretation may vary. >> >> but the coefficients for the zero weighted elements are NA, however. >> >> What is the reason for the warning? Shall I be concerned? >> >> In my actual data about 25% of the genes are weighted zero. >> >> This is my sessionInfo(): >> >>> sessionInfo() >> R version 2.7.0 (2008-04-22) >> i386-pc-mingw32 >> >> locale: >> LC_COLLATE=English_United Kingdom.1252;LC_CTYPE=English_United >> Kingdom.1252;LC_MONETARY=English_United >> Kingdom.1252;LC_NUMERIC=C;LC_TIME=English_United Kingdom.1252 >> >> attached base packages: >> [1] stats graphics grDevices datasets tcltk utils >> methods base >> >> other attached packages: >> [1] limma_2.14.0 svSocket_0.9-5 svIO_0.9-5 R2HTML_1.59 >> svMisc_0.9-5 svIDE_0.9-5 >> >> loaded via a namespace (and not attached): >> [1] tools_2.7.0 >>> >> >> >> Jose >> >> -- >> Dr. Jose I. de las Heras Email: J.delasHeras at ed.ac.uk >> The Wellcome Trust Centre for Cell Biology Phone: +44 (0)131 6513374 >> Institute for Cell & Molecular Biology Fax: +44 (0)131 6507360 >> Swann Building, Mayfield Road >> University of Edinburgh >> Edinburgh EH9 3JR >> UK >> >> -- >> The University of Edinburgh is a charitable body, registered in >> Scotland, with registration number SC005336. -- Dr. Jose I. de las Heras Email: J.delasHeras at ed.ac.uk The Wellcome Trust Centre for Cell Biology Phone: +44 (0)131 6513374 Institute for Cell & Molecular Biology Fax: +44 (0)131 6507360 Swann Building, Mayfield Road University of Edinburgh Edinburgh EH9 3JR UK -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.