biosimulators/Biosimulators_GillesPy2

Algorithms/solvers to support and KiSAO ids

jonrkarr opened this issue · 4 comments

jonrkarr commented

Solvers to initially support

KISAO terms exist for each of these algorithms and each of their parameters

  • gillespy2.solvers.numpy.ode_solver.ODESolver (integrator=lsoda): KISAO:0000088
    • atol (float, default=1e-12): KISAO:0000211: absolute tolerance
    • rtol (float, default=1e-6): KISAO:0000209: relative tolerance
    • lband (int, default=None): KISAO:0000480: lower half bandwith
    • uband (int, default=None): KISAO:0000479: upper half bandwith
    • nsteps (int, default=500): KISAO:0000415: maximum number of steps
    • first_step (float, default=0.0): KISAO:0000483: initial step size
    • min_step (float, default=0.0): KISAO:0000485: minimum step size
    • max_step (float, default=inf): KISAO:0000467: maximum step size
    • max_order_ns (int, default=12) : KISAO:0000219: maximum non-stiff order (Adams order)
    • max_order_s (int, default=5) : KISAO:0000220: maximum stiff order (BDF order)
  • gillespy2.solvers.cpp.ssa_c_solver.SSACSolver: KISAO:0000029: SSA
    • seed (int, default=None): KISAO:0000488
  • gillespy2.solvers.numpy.tau_leaping_solver.TauLeapingSolver: KISAO:0000039: tau-leaping
    • seed (int, default=None): KISAO:0000488
    • tau_tol (float, default=0.03): KISAO:0000228: epsilon

Additional solvers to support

These require additional KiSAO terms. We should discuss with Anna Zhukova how they would prefer to represent vode/adams and vode/bdf within KiSAO. Same for hybrid-tau/RK45, hybrid-tau/RK23, etc.

I'm also not sure what the TauHybridSolver is. Unfortunately, GillesPy2 doesn't have much documentation or references. I think this might be slow-scale tau-leaping (10.1016/j.cma.2008.02.024). We need to ask Brian Drawert.

  • gillespy2.solvers.numpy.ode_solver.ODESolver (integrator=dopri5): KISAO:0000087
    • atol (float, 1e-12): KISAO:0000211
    • rtol (float, 1e-6): KISAO:0000209
    • nsteps (int, 500): KISAO:0000415
    • first_step (float, 0): KISAO:0000483
    • max_step (float, inf): KISAO:0000467
    • safety (float, 0.9) : Safety factor on new step selection
    • ifactor (float, 10): Maximum factor to increase/decrease step size by in one step (description appears to be incorrect in SciPy)
    • dfactor (float, 0.2): Minimum factor to increase/decrease step size by in one step
    • beta (float, 0): Beta parameter for stabilised step size control
  • gillespy2.solvers.numpy.ode_solver.ODESolver (integrator=dop835): KISAO:0000436
    • Same parameters as dopri5
  • gillespy2.solvers.numpy.ode_solver.ODESolver (integrator=vode)
    • method (string, default=adams, range=['adams', 'bdf']): KISAO_0000475
    • with_jacobian (bool, default=false):
    • atol (float, default=1e-12): KISAO:0000211
    • rtol (float, default=1e-6): KISAO:0000209
    • lband (int, default=None): KISAO:0000480
    • uband (int, default=None): KISAO:0000479
    • order (int, default=12): KISAO:0000484
    • nsteps (int, default=500): KISAO:0000415
    • max_step (float, default=inf): KISAO:0000467
    • min_step (float, default=0.0): KISAO:0000485
    • first_step (float, default=0.0): KISAO:0000483
  • gillespy2.solvers.numpy.ode_solver.ODESolver (integrator=zvode)
    • Same parameters as vode
  • gillespy2.solvers.numpy.tau_hybrid_solver.TauHybridSolver: maybe KISAO:0000028 , need to ask developers
    • seed (int, default=None): KISAO:0000488
    • tau_tol (float, default=0.03): KISAO:0000228
    • integrator (string, default=lsoda, range=['RK45’, ‘RK23’, ‘Radau’, ‘BDF’, and ‘LSODA’]): KISAO_0000475

Solvers to potentially ignore

These solvers provide the same functionality as solvers above. We can ask Brian Drawert whether we should be using these, or the implementations above.

  • gillespy2.solvers.stochkit.stochkit_solvers.StochKitSolver (algorithm=ssa)
    • seed: KISAO:0000488
  • gillespy2.solvers.stochkit.stochkit_solvers.StochKitSolver (algorithm=tau_leaping)
    • seed: KISAO:0000488
    • ­epsilon
: KISAO:0000228
    • ­threshold (minimum reactions 
per 
leap): KISAO:0000230
  • gillespy2.solvers.stochkit.stochkit_solvers.StochKitODESolver
    Unclear whether this works and, if so, what the parameters are

Solvers to ignore

  • Identical to other solvers
    • gillespy2.solvers.cpp.variable_ssa_c_solver.VariableSSACSolver
      From release notes: "This solver works identically to SSACSolver, but is optimized for multiple trajectories using variable input values for species and Parameters."
    • gillespy2.solvers.numpy.ssa_solver.NumPySSASolver
      I assume this is equivalent to the C solver, but slower
  • Deprecated
    • gillespy2.solvers.cython.cython_ssa_solver.CythonSSASolver
    • gillespy2.solvers.numpy.basic_ode_solver.BasicODESolver
    • gillespy2.solvers.numpy.basic_tau_hybrid_solver.BasicTauHybridSolver
    • gillespy2.solvers.numpy.basic_tau_leaping_solver.BasicTauLeapingSolver

More documentation

https://gillespy2.github.io/GillesPy2/docs/build/html/classes/gillespy2.solvers.html
https://docs.scipy.org/doc/scipy/reference/generated/scipy.integrate.ode.html
https://docs.scipy.org/doc/scipy/reference/generated/scipy.integrate.solve_ivp.html

jonrkarr commented

I implemented this mapping in biosimulators_gillespy2.core.

Todo

  • request KISAO terms
  • test
jonrkarr commented

Submitted requests for new KISAO terms

  • 3 algorithms
  • 5 algorithm parameters

Ticket status: https://sourceforge.net/p/kisao/feature-requests/

jonrkarr commented

In the dev-kisao branch, I applied the newly requested KiSAO terms. This can be merged once the terms are incorporated into KiSAO.

jonrkarr commented

The above is implemented