[SED-ML-discuss] SED-ML roadmap

[Nicolas Le n. <le...@eb...>]

[I forward this discussion on the list with Frank's consent]

 >> Frank Bergmann wrote:
>>> - of special importance for many will be the interaction with external
>>> data sets. The SED-ML specification needs to specify how this
>>> interaction takes place. It should at least say the formats expected,
>>> probably SBRML and CSV for now. Right now the schema depicted in Fig
>>> 28, only lists measurement data linked to the Output class. This would
>>> be wasted potential. Think of parameter fitting experiments ... or
>>> think of time course simulation experiment defined over the same time
>>> range as given in the referenced data set. That is why the list of
>>> MeasurementData was described at the SED-ML level from the beginning.

 > Nicolas Le Novere wrote:
>> I entirely agree that the interaction with measurement data is missing.
>> And
>> that is why I was against it straight for the beginning. But now, 2 years
> after, I
>> even think this is not part of SED-ML at all. The parametrisation of a
> model is
>> not part of the simulation description. In certain case, simulations may
> take
>> part of some parametrisation determination procedures.
>> But describing those in SED-ML is a daunting challenge. In any case, it is
> not
>> part of v1. Again version 1, as agreed from day 1, will describe only the
> single
>> timecourse. Start small, get support, extend. That's the SBML philosophy.
>> And that works. We just have to be sure to have faster iteration than SBML
> :-
>> )
> 
> You've hinted at this many times now, also with your references to SBML L4.
> It would really be great to see your road map. 

I don't have a particular road-map for SED-ML, except that the first 
version was meant to encode the simple time-course. And in several of the 
past meetings, there was a relative consensus that the next important step 
was to tackle procedures such as parameter scans. A roadmap is indeed a 
good idea. We should put something like that on the website. Let's 
tentatively start:

Level 1: Single simulation procedures
     Version 1: Single timecourse
     Version 2: add ? [steady-state ...]

Level 2: Multiple independent simulations
     Version 1: Parameter scan
     Version 2: add ?

Level 3: Multiple linked simulations
    By that I mean simulation systems where a simulation depends, at 
run-time, on another simulation running concurrently.

The Version would supersede previous versions. The Levels would develop 
independently. We coul  But a Level N would be built on the Level N-1. The 
particular Version of Level N-1 to be used in a Version of Level N would be 
specified in its specification.

>>> -  AnySimulation: This type will never be exchangeable ... since the
>>> simulation experiment is not defined, it cannot be exchanged! The
>>> SED-ML standard already allows the specification of ANY, that is,
>>> PROPRIETARY simulation experiments through custom annotations. We
>>> should aim for people to use the standard classes, otherwise this
>>> project is doomed from the start ...
>> I totally agree.


-- 
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nle...@ho...(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere