We ran into some serious problem with our R/Bioconductor pipelines for Affymetrix gene expression arrays at arrayanalysis.org (and some others in the developer version) after the last R update from (2.13 -> 2.14)
First of all we used to use Rshell to run these procedures. Somehow that does no longer work. When run from Rshell Bioconductor is not able to read the .cel files. We changed to another shell. That helped.
But we also have problems with core Bioconductor packages which a.o. leads to annotations being wrong. The release notes mention something related to package environment changes, we are still investigating that.
Did anyone experience the same problems and find a generic solution or specific solutions for specific packages?
In any case this should serve as a warning. If you regularly run Bioconductor procedures and recently updated R do double check the output!
Update
The code used can be downloaded from arrayanalysis.org (click the yellow tab that says download source).
We try to tackle problems one by one.
First we were not able to run the pipeline from the server at all. That was solved by moving away from Rshell as mentioned above.
Next thing is that we need the right annotations to build the genesets for RMA. When the custom cdf's are not yet installed, it should get them, and that doesn't work. If we run the script directly from R, (so not as php invoked server code) all seems to go well. As soon as the custom cdf's are installed everything works and if we skip the step loading these cdfs the original cdf works fine. (We were told by another group that automatically getting the cdf's also causes problems for GCRMA where it tries to reinstall custom cdf's that are already there).
I am traveling. But I got an email explaining that the problem with standard cdf's after a custom cdf failed to load (I described that as a second problem earlier) was in fact a regular bug and not related to the update.
Apart from that there also are downstream procedures which don't work at all, or provide erroneous output. But we are not there yet.
What versions were involved? Was this 2.13 -> 2.14?
Yes, that is right correct.
Can you provide some more information, like a simple reproducible example? Sometimes creating the simple example helps to understand what the problem is.
Downloading the source code isn't good enough for a reproducible example -- what specifically am I supposed to do to trigger the errors you're seeing? When you say "When the custom cdf's are not yet installed, it should get them, and that doesn't work" can you give an example of the code in that part of the script, show the R commands and the errors that they produce? I'm not sure that Biostar is the right forum for this sort of trouble-shooting; have you thought of posting to the Bioconductor mailing list?
Martin you are right. I will try to get the isolated code part that causes the problems for the installation of the custom cdf files. We really have no clue what part of the code causes the problems with the default cdf as we can only see that from the output. The strange thing is both only happen when invoked from php on the server. I am traveling to New York at the moment. So getting that might take a while. My question was not meant to trouble-shoot our specific code, just to hear other experiences. Chris Miller's answer was very helpful in that respect.
@Chris: sessionInfo() is also a good idea to include when asking R-related questions.