pkident
An R package for structural and practical identifiability of PK/PD ODE models built on rxode2: which parameters can be determined from given outputs and doses, how precisely a sampling design would estimate them, and whether likelihood profiles of nlmixr2 fits are bounded or flat.
What it answers
- Structure: with continuous, noise-free observation of the chosen outputs after known doses, which parameters, or combinations such as V/F, can be determined at all.
- Design: with given sampling times, doses and assay error, how precisely each parameter would be estimated, and which added measurement, sampling time or dose level would help.
- Fit: in an nlmixr2 fit, which parameters the data determine, and whether a flat likelihood profile is a property of the model or of the data.
Installation
remotes::install_github("DaikiKumakura/pkident")pkident needs rxode2 and symengine; classify_profiles() also needs nlmixr2est and nlmixr2extra.
Status
Ten reference models with expected results fixed from the literature or by derivation are reproduced, and two independent structural methods agree at every point. Expected standard errors agreed with the spread of repeated estimates on simulated data. R CMD check passes on Linux and Windows in CI (October 5, 2026). Decisions are local; population designs are not covered.