[jcschaff]

vcml_to_ode_input couldn't convert a simulation in a rule-based application. VCell expands reaction rules into a reaction network with BioNetGen, and BioNetGen runs BNG2.pl from a VCell installation, which libvcell doesn't have. The failure surfaced as required System property "vcell.installDir" not defined.

It also regenerated every application's math (BioModel.updateAll), so a single rule-based application broke the input of every other simulation in the same document.

Change

• Only the simulation's own application is regenerated.
• A rule-based application uses the math stored in the document, with a warning. That's the case where VCell's NetworkTransformer applies, i.e. the application has enabled reaction rules. The stored math is the network VCell generated when the model was saved, and it's what the VCell server runs for a saved simulation.
• The success message says when stored math was used: "<input file>; stored math (rule-based application: BioNetGen, which would regenerate its reaction network, is not available in libvcell)". Input file names have no ;, so message.split(";")[0] is the name. A regenerated simulation's message is just the name, as in 0.0.20. In Java, SolverUtils.vcmlToOdeInput returns an OdeInput(inputFile, usedStoredMath) record.

How much it matters, and whether the stored math is right

• Scale: in a 2,730-model VCell corpus, 109 of the 1,411 models with ODE simulations (8%) have reaction rules, and every one stores math. Before this change, all of them failed.

• Checked against BioNetGen 2.9.3 (PyPI bionetgen, as a stand-in VCell installation) by regenerating 25 of them and comparing solver input:

  • The networks are the same in all 25: same equations, same expressions.
  • 8 give identical input.
  • The others differ only by Avogadro's number. VCell has since changed it from 6.02e23 to 6.02214179e23, a 0.036% change in some rates.

  So the stored math reproduces the model as VCell computed it when it was saved. Very sensitive models amplify that 0.036% difference, so a result from stored math is "the model as saved", not "regenerated by today's VCell". That's what the message note is for.

Tests

• Java: a test on public BioModel 100992568 ("CompartmentTransport", one reaction rule, downloaded anonymously from the VCell API) checks the input and usedStoredMath. SolverEntrypointsTest gives 23/23.
• Python: a test on the same model, through the native library, checks the message note and solves the input with pyvcell-odesolver. 28/28 pass locally on a fresh native build.
• MainRecorder exercises the path for native-image.
• make check is clean.

Follow-ups

• pyvcell's simulate_ode_vcml already takes the name before the ; (feat: run non-spatial simulations on VCell's Sundials ODE solver (simulate_ode) pyvcell#65).
• vcml_to_finite_volume_input still regenerates every application. It has the same problem when a rule-based application is in the document; that's for a separate change.
• pyproject.toml is at 0.0.21.

🤖 Generated with Claude Code