Re: [Miase-discuss] sed-ml library?

[Richard A. <ra...@st...>]

Hi all,
   Thanks to Dagmar's suggestions I've set up a Java project on
sourceforge for writing a library to support the specs. This uses JaxB
to parse the schema file and autogenerate classes, plan to add an API
on top of that. In order to get JaxB to work, there are a few small
changes needed in the schema, for example making some local element
declarations global, I don't think the semantics are changed.

   The project is at

     http://miase.svn.sourceforge.net/viewvc/miase/sed-ml/jlibsedml/

   Any input on API design etc very welcome!
   Dagmar suggested further implementation issues to be discussed on
the sourceforge miase bugzilla.

   Just working with this now there seem to be quite a lot of issues to address
in order to achieve an implementation. Here are some,

   A Schema issues:
   --------------
      1.  According to the spec SEDBase is a superclass of all the main
classes, yest in the schema only SedML class inherits from it. This
means that notes and annotation cannot be added to other elements,
e.g., uniform time course, dataset, datagenerator etc.,

      2. Most if not all top-level elements have an 'id' attribute,
would it make sense to put the 'id' field in SEDBase?

      3. If more than one simulation can be in the listOfSimulations,
then I think
        UniformTimeCourse needs a 'max=unbounded' attribute in its declaration,
        else ListOfSimulations only contains a single element.

   B)    Coordination with model
   ----------
     PRevious posts have raised this issue, is there any progress on
this? There is a lot of validation that can be done just with the
sedml file, but if we want to check that a parameter or variable name
exists, or the xpath is correct,
   then we need access to the model. It seems to me that there are
different scenarios:

        1) A model is referenced to a validated public repository e.g.,
BioModels
           In this case the library can download the model. This seems
the  easiest solution, as revisions of Biomodels are versioned (btw,
would a version attribute  for the model element be useful?) and so a
tool can make decisions on whether the model is correct or not.


         2) A model is not yet in the public domain, but for example
the client wants to share  simulation info with a collaborator. In
this case probably the model and  the sedml file would have to sent
together in a zip file to preserve
   relative file paths, but this leaves quite an onus on the user or
tool to keep the two files coordinated.

         3) Could a model file be optionally  embedded in a sedml file?
   This gets round the need to keep sedml independent of sbml, also a
single file gets round the file coordination issue,  but it raises new
problems of keeping in synch possibly multiple copies of a model. But
for non-curated models that must be an issue anyway?

C Validation requirements
--------------------------

   VAlidation against the schema offers one level of validation, but  
there are many more assertions to validate, here are
some below, would be good to have a list of these somewhere to
construct a test-suite.
Validation tasks below with an asterisk are ones needing access to the model:


   List of simulations (schema needs to be altered to hold 'max=unbounded'?)
     Unique simulation ids
1) Uniform time course -
     outputstart time < output end time
     outputst time >= initial time
     algorithm = valid Kisao id

2) List of Models
       how to check existence of model? *
       source = accessible url ( e.g., BioModelsDB)* or reference to
previously defined model
       No duplicate model IDs
       List of changes -> Check parameter name exists *

3) List of tasks

        One task must be baseline?

        model reference must refer to defined model id
        simulation reference must refer to defined simulation id
        No duplicate task IDs

4) list of data generators
      unique ids
      variable ID exists in model, target path exists in model
      task ref refers to valid task reference

5) List of outputs
        Plot ID is unique
        x and y data references refer to defined dataGenerator ID
        If listOfOuputs present, at least one of plot2d. plot3d, or
report is present.

Richard

Dr Richard Adams
Senior Software Developer,
Computational Systems Biology Group,
University of Edinburgh
Tel: 0131 650 8281/8285
email : ric...@ed...


-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.