Skip to contents

Run a simulation against an nlmixr-format model with rxode2

run_sim_nlmixr.Rd

Internal companion to run_sim(); called when the input model is a pharmpy nlmixr-backend model. Uses rxode2::rxSolve() directly so we can avoid the pharmpy-driven nlmixr fitting/simulation path (which requires the Python pyreadr package).

Usage

run_sim_nlmixr(
  fit = NULL,
  data = NULL,
  model = NULL,
  id = irxutils::get_random_id("sim_"),
  path = NULL,
  n_iterations = 1,
  variables = NULL,
  add_pk_variables = FALSE,
  output_file = "simtab",
  seed = 12345,
  verbose = TRUE,
  model_code = NULL
)

Arguments

fit

a Pharmpy modelfit object.

data

a NONMEM-format data.frame to use as the simulation dataset. Typically the output of create_sim_dataset(). If NULL, the dataset attached to model is used as-is.

model

either a Pharmpy model object, or a filename (for a model with NONMEM model code). If the latter, run_sim() will attempt to load the model into Pharmpy first.

id

base run id (default a random sim_*). NONMEM only: nlmixr2 simulations are solved in memory and write no run folders, so id does nothing there. Each regimen is run in its own subfolder id/regimen_<i> (<i> = 1-based regimen index), so regimens don't overwrite each other's output. Under uncertainty_engine = "replicates" each draw gets a folder of its own too, id/uncertainty_<r>/regimen_<i> (<r> = 1-based replicate index), so every replicate's NONMEM artifacts can be inspected afterwards and concurrent replicates cannot clobber each other.

path

ignored for the nlmixr2 backend: simulations run via rxode2::rxSolve() and create no NONMEM-style run folders. Accepted only to keep the signature aligned with run_sim().

n_iterations

number of iterations of the entire simulation to perform. The dataset for the simulation will stay the same between each iterations.

variables

vector of variables to output. If NULL, will output default variables c("ID", "TIME", "DV", "EVID", "PRED") as well as all variables declared in the NONMEM code.

add_pk_variables

calculate basic PK variables: CMAX_OBS, TMAX_OBS, CMIN_OBS, and (when CL is in the output table) AUC_SS. AUC_SS is derived as the last dose in the simulation dataset divided by CL.

output_file

TODO

seed

TODO

verbose

verbose output?

model_code

pre-rendered nlmixr2/rxode2 model code (character), used instead of extracting it from model. This is what lets a simulation run in a worker process: a Pharmpy model is a Python (reticulate) object and cannot cross a process boundary, but its rendered code can. When supplied, model/fit are not needed and data is required.

Details

Returns a data.frame in the same shape as the NONMEM-side simulation output (ID, TIME, DV, IPRED, PRED, EVID, plus declared variables and a regimen_label column), so downstream example code that plots simulation results works unchanged.

PRED is the population prediction, obtained from a second solve with the between-subject random effects zeroed (rxode2::zeroRe()) — rxSolve cannot emit IPRED and PRED in one call. If that solve is unavailable (older rxode2 without zeroRe()) or fails, PRED falls back to IPRED.