Deal with KINSOL used as the steady state algorithm.

[luciansmith]

Every biomodels that uses steady state uses the KISAO term 282 ('KINSOL'), and the checker returns this error:

NotImplementedError: Algorithm substitution for "KINSOL" (http://www.biomodels.net/kisao/KISAO#KISAO_0000282) for policy SAME_VARIABLES is not supported.

Now, if a model has an algebraic rule, it must use the KINSOL solver to find the steady state. But if it doesn't (and no biomodels does), other solvers can also be used.

There are two ways we can work with this, and perhaps both should be done:

1. Let the algorithm substitution algorithm have access to the referenced model. If there are no algebraic rules present, allow substitution of a different solver.

2. Change the biomodels SEDML files to use a more permissive steady state solver, since we know there aren't algebraic rules.

Any opinions?

The biomodels in question are:

final/BIOMD0000000498/BIOMD0000000498.sedml:
final/BIOMD0000000500/MODEL1311130001_highcoop.sedml:
final/BIOMD0000000500/MODEL1311130001_lowcoop.sedml:
final/BIOMD0000000501/MODEL1311130000_lowcoop.sedml:
final/BIOMD0000000574/MODEL1405060000_CaMKIIeffect.sedml:
final/BIOMD0000000637/Bush2016-Simplified-Carrousel-model-of-GPCR.sedml:
final/BIOMD0000000638/Bush2016-Extended-Carrousel-model-of-GPCR-RGS.sedml:
final/BIOMD0000000643/ARPP-16_Layer1_mutualInhibitions.sedml:
final/BIOMD0000000644/ARPP-16_Layer1and2_mutualInhibitions_PKAinhibitsMAST3.sedml:
final/BIOMD0000000645/ARPP-16_Layer1and2and3_mutualInhibitions_PKAinhibitsMAST3_dominantNegative.sedml:
final/BIOMD0000000803/Park2019.sedml:
final/BIOMD0000000847/Adams2019.sedml:
final/BIOMD0000000865/Nikolaev2019.sedml:
final/BIOMD0000000866/Simon2019.sedml:
final/BIOMD0000000868/simon2019_model2.sedml:
final/BIOMD0000000869/Simon2019_model3.sedml:
final/BIOMD0000000870/Simon2019_model4.sedml:
final/BIOMD0000000871/Simon2019_Figure3b.sedml:
final/BIOMD0000000881/Kogan2013.sedml:
final/BIOMD0000000919/Ledzewicz2013.sedml:
final/BIOMD0000000927/Grigolon2018_fig1b.sedml:

[jonrkarr]

The algorithm substitution is intended to be in reference to a particular simulation tool.

My recollection is that COPASI didn't originally export any SED-ML that needed KINSOL. Since there was no SED-ML with KINSOL, I didn't deal with it. I believe tellurium is the only case where we've mapped a KISAO term to a steady state method. All of the infrastructure is in place to do the same for other methods and tools. In this case, we'd ideally expose KINSOL through COPASI, or whatever COPASI implements. I think I might just need a little help from Frank to know what the appropriate COPASI syntax is.

Algorithm substitution might also need to be adjusted. Also relatively easy to do.

I think this summarizes the status for the interfaces to SBML tools:

• AMICI: no method to map
• COPASI: need a little help from Frank
• GillesPy2: no method to map
• LibSBMLSim: no method to map
• PySCeS: Could be mapped
• tellurium/libRoadRunner: mapped
• VCell: Not sure what the status is

[luciansmith]

Well, since it's COPASI itself that is exporting this SED-ML, one presumes that their normal steady state solver is KINSOL.

What do you think is best for Biomodels? COPASI is actually using KINSOL, but there are simulation tools that don't support it. It would be easy enough to swap out 282 for 407 (the generic 'a steady state algorithm') or 408 (the Newton method and parent of 282) when there aren't algebraic rules (which is, as of now, 100% of the time).

[jonrkarr]

I think the ideal is:

• Easy: If needed, change KiSAO or libKisao to recognize the similarity between nleq2 (tellurium) and KINSOL
• Fairy simple: Make a few changes to the COPASI interface to make this work -- see Expose static and steady-state analyses through SED-ML biosimulators/Biosimulators_COPASI#45 (has been open for > 1 year)
  • Fairly easy: Edit Python code
  • Easy: Annotate the capabilities of COPASI for this
  • Easy: Release new version of the COPASI interface
• Make similar change to tellurium interface
• Ideally, make similar changes to the PySCeS interface
• Encourage Ion to do the same for VCell

[hsauro]

Note roadrunner supports kinsol as of 4 months ago?

[luciansmith]

Oh, hey, excellent. Does that mean that we could support algebraic rules if we could store them? Or is that something different?

[jonrkarr]

Added tellurium/KINSOL to todo list above

[hsauro]

That's different, I put that into the next grant if we get it.