Skip to content

[Bioc-devel] Help with class lost after subsetting.

4 messages · Charles Plessy, Martin Morgan, Hervé Pagès

Original
Charles Plessy
# 
Dear Bioconductor developers,

In the CAGEr package, I created a "CAGEexp" class that extends
"MultiAssayExperiment" without adding new slots, in order to define generic
functions that require CAGEr-specific contents in the colData slot.

Unfortunately, when run in the development branch of Bioconductor,
the CAGEexp objects lose their class when they are subsetted.  Here
is an example:
A CAGEexp object of 4 listed
(...)
A MultiAssayExperiment object of 4 listed
(...)

This breaks examples in the package, as well as existing code.

I am lost on how to troubleshoot this.  May I ask for your help ?

Best regards,
Martin Morgan
#
On 08/28/2018 12:19 AM, Charles Plessy wrote:
I debugged this using first `selectMethod("[", "MultiAssayExperiment")` 
and then `showMethod()` / `selectMethod()` to arrive at 
`subsetByColData,MultiAssayExperiment,ANY-method`.

The problem is that this line

https://github.com/waldronlab/MultiAssayExperiment/blob/master/R/subsetBy-methods.R#L261

returns a MultiAssayExperiment; what it should do is probably closer to 
the 'copy constructor' functionality of `initialize()`, along the lines of

   initialize(x, ExperimentList = ..., )

This could be opened as an issue on the MultiAssayExperiment github 
repository; maybe Herve or Michael or others might comment on the best 
implementation.

Martin
Hervé Pagès
#
On 08/27/2018 11:01 PM, Martin Morgan wrote:
Yep. Personally I tend to prefer BiocGenerics:::replaceSlots()
over initialize() because the former can be called with check=FALSE
in order to skip a possibly expensive validation. So:

     BiocGenerics:::replaceSlots(x
         ExperimentList = harmon[["experiments"]],
         colData = harmon[["colData"]],
         sampleMap = harmon[["sampleMap"]],
         metadata = metadata(x),
         check = FALSE)

If you know that the replacement values are valid (because of the way
you prepared them), then validation should not be needed.

Also when only **some** of the slots are updated (which is not the
case in the above example where all the slots are being replaced),
I find that the use of initialize() is misleading from a readability
point of view.

See https://stat.ethz.ch/pipermail/bioc-devel/2017-September/011496.html
for a discussion about this about 1 year ago.

H.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Hervé Pagès
    

    

      
      #
    

  


  

      
On 08/28/2018 11:46 AM, Herv? Pag?s wrote:

            

      
      
      
    

        
ERRATA: It seems that MultiAssayExperiment have one more slot,
the "drops" slot, that the code above does not modify so this would
be one more reason IMO to use BiocGenerics:::replaceSlots() instead
of something like initialize(x, ExperimentList = ..., ) or
new(class(x),ExperimentList = ..., ).

H.