[matthiaskoenig]

Regenerates v1/simulations.tsv of five problems whose simulations are not the model at the nominal parameters, see #278 (comment with the details), #181 and AMICI-dev/AMICI#3503. Closes #181.

Only the simulation column changes (and, for Fujita, the condition ids); every row stays where it was, i.e. the tables keep their columns and are row by row the measurements.

problem	before	after
Fujita_SciSignal2010	15 of 144 simulations off by up to 0.5 % (abs 0.0045); conditions model1_data*, which the problem renamed to condition_step_*	conditions of the measurements
Sneyd_PNAS2002	28 of 135 off, all in Ca_dose_response__5 and __6 (up to 0.027)
Zheng_PNAS2012	the measurements (#181)
Perelson_Science1996	not the model at the nominal parameters (V(0.105) = 1.979e6, the model gives 1.859e6)
Chen_MSB2009	5 of 120 off by up to 0.28 % (AKT_PP, model1_data4, t = 1950: 17.7718, the model gives 17.7228)	the simulations of AMICI 1.2.0 (simulate_petab, tolerances 1e-16 / 1e-12), which sbmlsim reproduces to 4e-6

Check: the log-likelihood of the tables with petab.v1.calculate.calculate_llh (petab 0.10.0; the simulations matched to the measurements by row, as the tables carry the values of the noise parameters, which calculate_llh compares as columns) against the reference of AMICI v1.2.0 (tests/benchmark_models/benchmark_models.yaml):

problem	llh before	llh after	AMICI
Fujita_SciSignal2010	52.840225	53.083767	53.083777
Sneyd_PNAS2002	178.504049	319.791777	319.791778
Zheng_PNAS2012	308.333621	278.333560	278.333533
Perelson_Science1996	-432.038550	-432.038550	none
Chen_MSB2009	-130669.488	-130591.753	-115700.119 (stale, AMICI 1.2.0 itself gives -130591.753 at tolerances from 1e-8 to 1e-16, see AMICI-dev/AMICI#3503)

The llh of Perelson_Science1996 hardly depends on the simulations, the nominal sd_task0_model0_perelson1_V is 1e5 on the log10 scale; its new simulations agree with an independent integration of model.xml with SciPy (solve_ivp, Radau, rtol 1e-12) to 2e-12 relative.

How: simulated with sbmlsim (libroadrunner, CVODE with absolute and relative tolerances 1e-14 and 1e-12) on the problems converted with petab.v2.petab1to2 at the nominal values of their parameter tables. The values may differ from an AMICI run in the last digits; src/python/simulate.py with AMICI should reproduce them within the usual tolerances.

[matthiaskoenig]

@dweindl Could you merge this or better recreate the tables with AMICI v1.2.0 to fix the values (using the simulate script). I would prefer to have the AMICI values here, so we can test that our simulator creates the same values (LLH).

[dilpath]

Hi Matthias, thanks for the contribution. I moved your commits over to #344

@dweindl Could you merge this or better recreate the tables with AMICI v1.2.0 to fix the values (using the simulate script). I would prefer to have the AMICI values here, so we can test that our simulator creates the same values (LLH).

Done in #344

[matthiaskoenig]

Perfect. Thanks.