[BUG] SED-ML files for native and SBML models have involve different numbers of models (archives are not equivalent)

[jonrkarr]

Example (12119723):

• Native: https://github.com/virtualcell/vcdb/blob/main/published/biomodel/omex/native/biomodel_12119723.omex
• SBML: https://github.com/virtualcell/vcdb/blob/main/published/biomodel/omex/sbml/biomodel_12119723.omex

The SED-ML for native has 1 model, 3 simulations, and 3 tasks.

The SED-ML for SBML has 3 models, 3 simulations, and 3 tasks. The SBML files used by the SED-ML are not identical.

Because the SBML files used by the SED-ML are not identical, the two versions of the COMBINE archives are not equivalent.

[moraru]

This is not a bug, it is by design, and is correct. The native format (VCML) encodes one or more variations of a model, each of which needs to be translated into a corresponding SBML model upon export. The tree structure of the VCell BioModel is roughly equivalent to a SEDML file, With the Application being the model and the Simulation being the task.

Therefore there will always be exactly 1 model when creating OMEX with VCML format but 1..n models when creating OMEX with SBML format. The total number of tasks however should always be identical.

[moraru]

@jonrkarr if you want to check equivalency you need to count the tasks and match the simulation names

[jonrkarr]

Understand that. The issue is that the VCML vs SBML tasks are different:

• The VCML OMEX involves 3 tasks for the same model (there are no model changes encoded into SED-ML)
• The SBML OMEX involves 3 tasks for 3 different models (the SBML files have different parameter values)

If the VCML tasks are intended to simulate different variants of a model, I think this should be encoded using SED-ML model changes, which is not presently the case. It sounds like this is what's happening:

• Either model changes are encoded into VCML (meaning they are hidden from SED-ML) or they are not correctly encoded into SED-ML
• Model changes encoded into SED-ML are potentially being ignored

From my perspective, this is unintuitive because it deviates from how SED-ML is intended to be used. In the case of other model languages (SBML-fbc, SBML-qual, BNGL, NeuroML/LEMS, Smoldyn, RBA XML, XPP ODE), I have used SED-ML to capture model changes, including when those languages have capabilities to capture model variants themselves. Encoding model changes into SED-ML is important to enabling us to use SED-ML as an abstraction for specifying changes to models in any format.