Update PETab to be model format agnostic

[luciansmith]

Which problem would you like to address? Please describe.
The CellML team is about to implement support for parameter estimation in OpenCor, and originally investigated PETab as the format to use. However, everything was SBML-specific, so instead they're considering the only other alternative, the SED-ML L1v4 more generified version of parameter estimation.

Describe the solution you would like
All the spec needs to do to change is to replace SBML-specific language with language that defined 'a well-defined model definition language', perhaps adding 'such as SBML or CellML'. And frankly, you could probably get away with saying 'SBML or CellML' specifically, but it would probably be nice to be just generally generic. There are a few other model languages out there like BNGL (for rule-based modeling) or Smoldyn (for spatial modeling) that are not (currently) compatible with SBML, but might be able to use PETab.

Describe alternatives you have considered
The main alternative is to have PETab remain SBML-specific, and let SED-ML's L1v4 definitions be the model-agnostic form. However, that means that it becomes much more difficult to translate between the two, or that interpreters much either choose one or the other, or spend more time becoming compatible with both. It would be more efficient to have only one standard going forward.

Additional context
I think the SED-ML editors would be happy to drop the SED-ML version of parameter estimation in favor of PETab if PETab was agnostic about its models. It would certainly reduce the amount of work necessary in the modeling community, which would be a good thing.

[matthiaskoenig]

Hi Lucian,
the following proposal addresses the issue and has been implemented for BNGL in pysb
#538
Should work for cellml right now.
Best Matthias

[luciansmith]

Aha! OK, thanks--@nickerso does this work for you all?

[dweindl]

What Matthias said.

If there are any other changes necessary to properly work with CellML, we can still adapt #538 accordingly.

[fbergmann]

Additional context
I think the SED-ML editors would be happy to drop the SED-ML version of parameter estimation in favor of PETab if PETab was agnostic about its models. It would certainly reduce the amount of work necessary in the modeling community, which would be a good thing.

I think it would be a mistake for the editors to remove PE from SED-ML. Remember, that PEtab only describes the PE setup, not what to do with the PE problem once you set it up. While that can be standardized as well it will be hard and will take time. Since there is no urgent need to remove things from SED-ML that could potentially be specified there, this should not be done.

On the specific issue of CellML it could probably be used in PEtab with the translation table that was introduced along with the language independent format. (That way you could target specific elements in subcomponents).