



## 23.6.2008, Brixen:

* Keep code in the weird fake Vignette and set eval=FALSE

* Drop the slot "state", it is not really useful. FH: Why is that?
  Seems to be useful for validity checking. We don't want to dig in
  other slots to get this information if we don't have to...

* Interactive links to various annotations (Doro knows what this is)

* use consistent wording for columns of the NChannelSet 
  (instead of replicate, treatment). E.g.: "assay"

* Can we drop the slot "batch"? It mostly seems to blow the object
  structure and complicate the code. There are two types of batch 
  effects: 
  - between sets of replicates / conditions (i.e. columns) - these 
    go into phenoData
  - between sets of plates (i.e. rows) - see next item

* We want a slot "plateData" which is a list of dataframes or 
  AnnotatedDataframes. #DONE#
  Each df has as many rows as plates, and as many columns as the 
  NChannelSet. One of the dfs, for example, can code for "batch". 
  And Florian wants to use "batch" in some of the diagnostic plots.

* The writeReport should strongly encourage users to provide the
  history of the object being shown in the form of the R script with
  which they created it (i.e. writeReport ask . #DONE#

* store normalization and summarization methods and use this
  information in the output. #DONE#

* check the enormous list of dependencies of the package, see which 
  ones we can remove. #DONE#

* People were confused on how to get the data out. This should be 
  explained in our vignette in a self-contained manner, since most 
  people don't know how an NChannelSet 


* subset method is broken (see Florian's mail of 30.4.08)

* summarizeReplicate is fine if there are no biologically 
  different assays and they are all just replicates; for any more 
  complex design, document how to use limma's lmFit and eBayes.

* standard deviations plot: wouldn't a standard error plot be more informative?