Simulate

Running a model

Simulate runs validated physics and chemistry models in an isolated worker with bounded time and memory, and records the model version and inputs with every result.

• In Setup, choose the model and change its inputs; units and allowed ranges are shown next to each. The model's description lists its assumptions, validity, known failure modes and citations.
• What to do: Run once, or Optimize to let Bayesian optimization run the model repeatedly (see Simulation in the BO loop).
• The result shows the values, curves and a quality check, with Download (CSV or JSON), Save as dataset and Open in Plot Studio.
• Past runs groups runs that belong together (compared sets, optimization loops, consecutive runs of one model). Reopen a run, compare 2–8 runs, or compare 2–10 runs with measured data on one graph.

Packages and limits

Quantum chemistry and molecular dynamics run on a separate chemistry worker with 4 GB of memory. A DFT or water MD run uses 5 runs, an xTB run 2, every other model 1 (see Runs).

Worked chemistry examples

Ready-made models

23 models. The design inputs are what Bayesian optimization searches when you choose Optimize; the objective is fixed by the model.

Fit to measured data

Fitting works for batch reaction kinetics, EIS Randles, the adsorption isotherm, fixed-bed breakthrough and custom_ode: choose 1–6 parameters with bounds and a measured column in the project (in the same units as the model), optionally a σ column. The fit uses 10–400 model evaluations and counts as one simulation. It reports the values with intervals, and the fitted curve can be saved as a dataset.

Writing your own model

When no ready-made model fits (an enzyme cascade, a reaction network, a dose–response), write the rate equations yourself as custom_ode, or plain formulas as custom_algebraic. AcquiLAB evaluates them with a whitelist of operations, never arbitrary code, and integrates them with SciPy's LSODA solver.

Full example: the creatinine sensing cascade (CA → CI → SOX)

The model behind the 4 October 2026 run in the Creatinine sensor project, with the literature constants used there. It passed AcquiLAB's checker together with the optimization settings below (total enzyme fixed at 3 µM, 20 runs). That run found an optimum near 1 : 1.3 : 1.7 (CA : CI : SOX), 0.0196 mM H₂O₂ at 5 min against 0.0186 mM for 1 : 1 : 1, with sarcosine oxidase as the rate-limiting step.

{
  "states": [
    {"name":"Cn",  "initial":0.1,  "unit":"mM"},   // creatinine
    {"name":"Cr",  "initial":0,    "unit":"mM"},   // creatine
    {"name":"Sar", "initial":0,    "unit":"mM"},   // sarcosine
    {"name":"O2",  "initial":0.25, "unit":"mM"},
    {"name":"H2O2","initial":0,    "unit":"mM"}],
  "parameters": [
    {"name":"kcat1","value":252, "unit":"1/s","source":"BRENDA, P. putida creatininase"},
    {"name":"Km1",  "value":26,  "unit":"mM", "source":"J Biochem 1979"},
    {"name":"Kcr",  "value":130, "unit":"mM", "source":"J Biochem 1979"},
    {"name":"Keq",  "value":1.3, "unit":"1",  "source":"assumed"},
    {"name":"kcat2","value":10.5,"unit":"1/s","source":"derived from 14 U/mg"},
    {"name":"Km2",  "value":1.33,"unit":"mM", "source":"BRENDA"},
    {"name":"kcat3","value":14,  "unit":"1/s","source":"J Appl Biol Sci 33:161"},
    {"name":"Km3",  "value":3.6, "unit":"mM", "source":"J Appl Biol Sci 33:161"},
    {"name":"KmO2", "value":0.1, "unit":"mM", "source":"assumed"},
    {"name":"E1","value":0.001,"unit":"mM","source":"assumed"},
    {"name":"E2","value":0.001,"unit":"mM","source":"assumed"},
    {"name":"E3","value":0.001,"unit":"mM","source":"assumed"}],
  "rates": {
    "Cn":  "-kcat1*E1*(Cn - Cr/Keq)/Km1/(1 + Cn/Km1 + Cr/Kcr)",
    "Cr":  "kcat1*E1*(Cn - Cr/Keq)/Km1/(1 + Cn/Km1 + Cr/Kcr) - kcat2*E2*Cr/(Km2 + Cr)",
    "Sar": "kcat2*E2*Cr/(Km2 + Cr) - kcat3*E3*Sar/(Km3 + Sar)*O2/(KmO2 + O2)",
    "O2":  "-kcat3*E3*Sar/(Km3 + Sar)*O2/(KmO2 + O2)",
    "H2O2":"kcat3*E3*Sar/(Km3 + Sar)*O2/(KmO2 + O2)"},
  "report": {"name":"H2O2 at 5 min","expression":"H2O2","unit":"mM"},
  "time":   {"end":5,"unit":"min","points":301}
}

// optimize the enzyme split at a fixed total of 3 µM
optimize    = [{"name":"E1","lower":0.0002,"upper":0.0026},
               {"name":"E2","lower":0.0002,"upper":0.0026},
               {"name":"E3","lower":0.0002,"upper":0.0026}]
constraints = [{"sum":["E1","E2","E3"],"equals":0.003}]
goal = "maximize", max_simulations = 20

Comments (//) are for reading only; leave them out of the real JSON.

Custom ODE library

Every row was written in AcquiLAB's format and passed its checker. Copy a rate, give the parameters values and sources, and you have a runnable model. Up to 8 states fit in one system, so rows can be combined (for example a cascade plus enzyme inactivation).

Formula library

custom_algebraic has parameters, named outputs (up to 8) and a report, with no states or time. Use it for closed-form relations, or as a fast objective inside an optimization loop. Non-integer powers are written as exp(n*log(...)), and x is reserved. A system holds at most 24 parameters, so split large formula sets.

Simulation in the BO loop

Choose a model, then What to do → Optimize. Bayesian optimization runs the model repeatedly, choosing each new set of inputs from the results so far. Settings stay as they are; only the inputs inside the search ranges change.

The loop page follows each run. Pause, resume, stop, or approve and skip single runs (a reason is optional). The loop reserves its whole run budget when it starts and gives back what it did not use.

Constraints are available for written models; a loop on a ready-made model searches its design inputs within the ranges you narrow.