Skip to contents

Benefits of Intermediates

intermediateFiles.Rmd

It may appear inconvenient to go through so many steps

SBtab -> .vf -> .c -> .so,

but something like this happens in most simulators anyway. Here it happens more explicitlty and we purposefully draw the users attention to these files. Each is an opportunity to verify correctness, add functionality, or modify.

Every step produces an output file that carries a distinct meaning:

  • the SBtab spreadsheet describes the biological model (so does SBML)
    • but with some consideration regarding systems (input/output model)
  • the .vf file represents the model as an ODE
    • it doesn’t have “reactions” anymore
  • the .c file is the code representation of the ODE model, intended for a specific solver suite.
    • it is not a general ODE anymore and can have C specific syntax (pow, ?:)

Each of these files can be checked by the user, their internal formats are not hidden; this can help with checking for errors and testing, resability, and interoperability between different software packages.

Note on VFGEN

The vfgen software can also create model files for both R and C and even two different solver libraries in C: cvode and GSL; but this is not all, vfgen can create models for many other langueges. It is really good!

The functions it creates are slightly different from ours.

We decided to write our own converter to better suit our needs, for several reasons:

  • vfgen creates one C function per vf-Function (xml-element), with the output’s name (the Name in .vf), e.g.:
    • double myModel_ABCoverSUM(double t, double y_[], void *par) {
  • ode.sh creates one output function that returns a vector
    • int myModel_func(double t, doubel y_[], double *func_, void *par){
    • i.e. vector valued output, like all the other functions
  • vfgen does not write a default initial conditions or default parameter function – these should be functions, because they can depend on
    • the constants in the model,
    • the previous parameters,
    • and even on the initial time \(t_0\), e.g.: double par12 = t>0 ? par3*par5 : par5;
  • the vfgen functions do not return an error code for NULL in-out-buffers (which is fine)
    • we use these error codes to probe the dimensionality of the model, without writing extra data structures
    • AKAR4cl_vf(0,NULL,NULL,NULL) returns 2, because AKAR4cl has two state variables; but 0 on success.

We can easily add more functionality to ode.sh because it is a stupid shell script, but not to vfgen (c++ is hard).