Hi Wolfgang, I upgraded to tilingArray 1.13.3 and R-2.5 and I also upgraded all the other packages as far as I could, but I'm afraid with no effect. The problem doesn't occur on Linux, everything works fine there. Thanks for your quick response and help. Greetings Julian Gehring ------- > sessionInfo() R version 2.5.0 Under development (unstable) (2007-02-26 r40806) i386-pc-mingw32 locale: LC_COLLATE=German_Germany.1252;LC_CTYPE=German_Germany.1252;LC_MONETAR Y=German_Germany.1252;LC_NUMERIC=C;LC_TIME=German_Germany.1252 attached base packages: [1] "splines" "grid" "tools" "stats" "graphics" "grDevices" [7] "utils" "datasets" "methods" "base" davidTiling GO tilingArray pixmap geneplotter annotate "1.0.4" "1.15.1" "1.13.3" "0.4-6" "1.12.0" "1.13.6" genefilter survival vsn strucchange sandwich zoo "1.13.8" "2.31" "2.0.15" "1.3-1" "2.0-1" "1.2-2" RColorBrewer affy affyio Biobase "0.2-3" "1.13.14" "1.3.3" "1.13.39" -----Original Message----- From: Wolfgang Huber [mailto:huber@ebi.ac.uk] Sent: Mittwoch, 28. Februar 2007 18:12 To: Julian Gehring Cc: bioconductor at stat.math.ethz.ch Subject: Re: [BioC] Normalization in davidTiling demo Hi Julian, I have heard about this problem now a few times from people that use R 2.4 and Bioconductor release 1.9 on Windows, but was never able to reproduce it. I have just rerun the vignettes "segmentation.Rnw" and "assessNorm.Rnw" from the tilingArray/inst/scripts directory on the system detailed below(*), with no problems. (Note that you don't need to type the commands from the PDF, but can use the functions 'Sweave' or 'Stangle' to process vignettes.) So, to nail this down: 1. Does your problem persist if you upgrade to tilingArray version 1.13.3? http://bioconductor.org/packages/2.0/bioc/html/tilingArray.html Since the version you used, there have been a number of updates in particular to the normalizeByReference function. 2. If yes, does it persist if you use the R devel (to be 2.5) snapshot build: http://www.stats.bris.ac.uk/R/bin/windows/base 3. If yes, does it also occur on Linux or Mac? Once we have the answer to these, we can see further. Thanks for pointing this out and helping to trace it down. Best wishes Wolfgang Huber EBI (*) > sessionInfo() R version 2.5.0 Under development (unstable) (2007-02-28 r40808) x86_64-unknown-linux-gnu locale: LC_CTYPE=en_GB.UTF-8;LC_NUMERIC=C;LC_TIME=en_GB.UTF-8;LC_COLLATE=en_GB .UTF-8;LC_MONETARY=en_GB.UTF-8;LC_MESSAGES=en_GB.UTF-8;LC_PAPER=en_GB. UTF-8;LC_NAME=C;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_GB.UTF-8 ;LC_IDENTIFICATION=C attached base packages: [1] "splines" "grid" "tools" "stats" "graphics" "grDevices" [7] "utils" "datasets" "methods" "base" other attached packages: Scerevisiaetilingprobe matchprobes davidTiling "1.1.0" "1.7.4" "1.2.1" GO tilingArray pixmap "1.15.1" "1.13.3" "0.4-6" geneplotter lattice annotate "1.13.7" "0.14-16" "1.13.6" genefilter survival vsn "1.13.8" "2.31" "2.0.16" strucchange sandwich zoo "1.3-1" "2.0-1" "1.2-2" RColorBrewer affy affyio "0.2-3" "1.13.14" "1.3.3" Biobase fortunes "1.13.39" "1.3-2"

Dear Julian, oh well. The non-linear optimisation routine L-BFGS-B that is used in vsn (which is used in normalizeByReference for between-array normalisation) can sometimes be a fit fiddly with its convergence criteria. It seems there are some OS- or compiler-specific differences. I am currently working on a completely revised version of vsn, and once the normalizeByReference function has been been updated, this problem may be solved. For the short term: - use Linux, or - note that vsn is only used for the between-array normalization; the main point of the normalizeByReference function is to adjust for probe effects. Hence you can call the function on each RNA-array separately, this will avoid the between-array normalization step but still do the same kind probe effect adjustment. You can then do the between-array normalization by some other means (e.g. just shift to the same median). Best wishes Wolfgang > Hi Wolfgang, > > I upgraded to tilingArray 1.13.3 and R-2.5 and I also > upgraded all the other packages as far as I could, but I'm > afraid with no effect. > The problem doesn't occur on Linux, everything works fine > there. > Thanks for your quick response and help. > > Greetings > Julian Gehring