I'm developing a new gene expression and survival analysis platform based on Bioconductor, but also including the custom annotation products from the BrainArray lab - http://brainarray.mbni.med.umich.edu - which could potentially become the new standard for some widely used microarray platforms. More use would also be made of the Gene Ontology, moving away from an emphasis on individual genes, whose expression values can sometimes be very noisy and greatly in error. Also many widely used statistical techniques are themselves to some extent flawed, leading to increased numerical instability and generating results which are not reproducible across different labs.

I ran into some severe problems just trying to install packages, encountering problems which had been reported for years, but had never been solved or even reproduced by the Bioconductor support team. It now appears that the major cause is various packages being out of sync in their version numbers on the local platform. Both Bioconductor and R have fairly elaborate mechanisms for ensuring version compatibility across packages, but apparently the checking is not 100%, leading in some cases to inconsistent package libraries and incomprehensible diagnostic messages.

I ended up manually downloading and installing key packages as listed in their core documentation to ensure that 
I had the right versions, a task made more difficult by Mac OS X being too helpful and automatically expanding archives which could then no longer be installed.

My personal feeling is that the version compatibility issue could warrant some basic redesign in Bioconductor

Alan J. Robinson   robin073@umn.edu

Hi Alan,

try to be more specific about the problem you are describing. In particular, the posting guide at http://www.bioconductor.org/help/support/posting-guide recommends 4 key points to follow among which I would suggest you to provide the following two:

• Provide a minimal, self-contained example (data and code) that someone else can cut-and-paste into a new R session to reproduce your problem.
• Include the output of sessionInfo(). If the example produces an error, provide the output of traceback() after the error occurs.

cheers,

robert.

Hi Alan -- Were you installing packages using biocLite() as described on the 'landing page' of each Bioconductor package, ensuring that you did not mix libraries from different R or Bioconductor versions? If so, then a first step to involving us in addressing your concerns is to provide a reproducible illustration of the problems you encountered. The bioc-devel mailing list may be a more appropriate forum for a discussion of strategies for improving package delivery (helping users address specific installation problems is appropriate for the current forum).

Obviously we cannot solve problems that we cannot reproduce; we do try to make installation and coordinated versioning more robust, the development of the BiocInstaller package being one recent manifestation of that, and R's transition from twice- to once-yearly releases and the difference in package versioning strategies adopted by R (most recent versions available, no coordination across packages) versus Bioconductor (distinct 'releases' of coordinated package) representing new or ongoing challenges. There are often complex trade-offs involved, not all of which are under our control.