Skills › Scientific Agent Skills › pkpd-modeling › software-ecosystem.md

software-ecosystem.md

64 lines · skills/pkpd-modeling/references/software-ecosystem.md at 5f2bc03

1# PK/PD software interfaces reviewed 2026-10-01

3Choose a tool by required model, diagnostics and verification evidence. Neither programming

4language nor vendor establishes suitability for a regulatory or clinical context. Bundled scripts

5are tested exploratory implementations, not a validated clinical platform.

PK/PD software interfaces reviewed 2026-10-01

Choose a tool by required model, diagnostics and verification evidence. Neither programming language nor vendor establishes suitability for a regulatory or clinical context. Bundled scripts are tested exploratory implementations, not a validated clinical platform.

Raw lines 1 to 6

7## Executed Python interfaces

9NumPy 2.5.3 and SciPy 1.18.1 were used for isolated tests. NumPy2+ provides `np.trapezoid`; it does

10not sort sample times. SciPy `least_squares`, `minimize`, `solve_ivp` and distribution functions

11supply numerical primitives; optimizer/solver success alone is not scientific validation.

13Pharmpy 2.2.0 (26 August 2026) requires Python>=3.12, excluding 3.14.1. Installed signatures and seven

14independent transformations were checked. Important changes: dataset indices begin at 1 since 2.0;

15`add_placebo_model` became `set_placebo_model` in 2.1; `set_unit` became `annotate_unit` in 2.2.

16Annotation declares a unit; `convert_unit` changes values. Inspect observation IDs because

17`get_observations` includes all DVIDs by default.

18[Changelog](https://pharmpy.github.io/latest/changelog.html).

20```python

21# Executed independently on Pharmpy 2.2.0's bundled model; no estimation engine required.

22import pharmpy.modeling as m

23base = m.load_example_model("pheno")

24tmdd = m.set_tmdd(base, type="qss")

25turnover = m.add_indirect_effect(base, expr="emax", prod=True)

26direct = m.set_direct_effect(base, expr="sigmoid")

27link = m.add_effect_compartment(base, expr="emax")

28size = m.add_allometry(base, allometric_variable="WGT", reference_value=70)

29transit = m.set_transit_compartments(base, n=3, keep_depot=True)

30nonlinear = m.set_michaelis_menten_elimination(base)

31```

33Transformations return new models; unassigned calls leave the original unchanged. A successful

34transformation is not a fitted/validated model. `set_tmdd` observation mapping is described in

35[TMDD](tmdd-and-biologics.md), and shrinkage's `sd` convention in [population PK](population-pk.md).

37The installed tools include run_allometry, run_amd, run_bootstrap, run_covsearch, run_estmethod,

38run_iivsearch, run_iovsearch, run_linearize, run_modelfit, run_modelrank, run_modelsearch,

39run_pdsearch, run_qa, run_retries, run_ruvsearch, run_simulation, run_structsearch, run_tool and

40run_vpc. Engine-backed runs require separately configured estimation software and were not executed.

41[Tool documentation](https://pharmpy.github.io/latest/tools.html).

Executed Python interfaces

NumPy 2.5.3 and SciPy 1.18.1 were used for isolated tests. NumPy2+ provides np.trapezoid; it does not sort sample times. SciPy least_squares, minimize, solve_ivp and distribution functions supply numerical primitives; optimizer/solver success alone is not scientific validation.

Pharmpy 2.2.0 (26 August 2026) requires Python>=3.12, excluding 3.14.1. Installed signatures and seven independent transformations were checked. Important changes: dataset indices begin at 1 since 2.0; add_placebo_model became set_placebo_model in 2.1; set_unit became annotate_unit in 2.2. Annotation declares a unit; convert_unit changes values. Inspect observation IDs because get_observations includes all DVIDs by default. Changelog.

# Executed independently on Pharmpy 2.2.0's bundled model; no estimation engine required.
import pharmpy.modeling as m
base = m.load_example_model("pheno")
tmdd = m.set_tmdd(base, type="qss")
turnover = m.add_indirect_effect(base, expr="emax", prod=True)
direct = m.set_direct_effect(base, expr="sigmoid")
link = m.add_effect_compartment(base, expr="emax")
size = m.add_allometry(base, allometric_variable="WGT", reference_value=70)
transit = m.set_transit_compartments(base, n=3, keep_depot=True)
nonlinear = m.set_michaelis_menten_elimination(base)

Transformations return new models; unassigned calls leave the original unchanged. A successful transformation is not a fitted/validated model. set_tmdd observation mapping is described in TMDD, and shrinkage's sd convention in population PK.

The installed tools include run_allometry, run_amd, run_bootstrap, run_covsearch, run_estmethod, run_iivsearch, run_iovsearch, run_linearize, run_modelfit, run_modelrank, run_modelsearch, run_pdsearch, run_qa, run_retries, run_ruvsearch, run_simulation, run_structsearch, run_tool and run_vpc. Engine-backed runs require separately configured estimation software and were not executed. Tool documentation.

Raw lines 7 to 42

43## Specialist tools reviewed through official documentation

45| Task | Tools and boundary |

46| --- | --- |

47| NLME | NONMEM 7.6, Monolix, nlmixr2; select estimation/diagnostics appropriate to data |

48| NCA | Phoenix WinNonlin, PKNCA; predefine conventions and qualify the implementation |

49| Simulation | rxode2, mrgsolve, Simulx; preserve event semantics and residual/IIV distinctions |

50| BE | PowerTOST for planning; validated replicate-design analysis for ABEL/RSABE |

51| PBPK | PK-Sim/MoBi 12.3, Simcyp, GastroPlus; context-specific verification |

53CRAN nlmixr2 currently reports 7.0.1 with R>=4.1 and rxode2>=5.0.0. Match companion-package

54versions; recent rxode2 serialization changes mean old saved objects may need conversion/refitting.

55[CRAN package](https://cran.r-project.org/web/packages/nlmixr2/index.html),

56[rxode2 news](https://nlmixr2.github.io/rxode2/news/index.html).

57NONMEM 7.6 includes ADVAN16/17 for delay equations; consult its installed manual for control-stream

58semantics. [Vendor overview](https://www.iconplc.com/solutions/technologies/nonmem).

60PKPy's [author repository](https://github.com/Gumgo91/PKPy) describes its Python PK workflow;

61`pkpy` returned 404 from PyPI during review, so do not assume that package name is installable there.

62`chi-drm`1.0.3, PINTS0.6.1 and lmfit1.3.4 metadata were checked, but no API examples or model fits

63for them were executed. They are optional alternatives, not dependencies of these scripts.

64See [source ledger](source-ledger.md) for evidence boundaries.

Specialist tools reviewed through official documentation

Task	Tools and boundary
NLME	NONMEM 7.6, Monolix, nlmixr2; select estimation/diagnostics appropriate to data
NCA	Phoenix WinNonlin, PKNCA; predefine conventions and qualify the implementation
Simulation	rxode2, mrgsolve, Simulx; preserve event semantics and residual/IIV distinctions
BE	PowerTOST for planning; validated replicate-design analysis for ABEL/RSABE
PBPK	PK-Sim/MoBi 12.3, Simcyp, GastroPlus; context-specific verification

CRAN nlmixr2 currently reports 7.0.1 with R>=4.1 and rxode2>=5.0.0. Match companion-package versions; recent rxode2 serialization changes mean old saved objects may need conversion/refitting. CRAN package, rxode2 news. NONMEM 7.6 includes ADVAN16/17 for delay equations; consult its installed manual for control-stream semantics. Vendor overview.

PKPy's author repository describes its Python PK workflow; pkpy returned 404 from PyPI during review, so do not assume that package name is installable there. chi-drm1.0.3, PINTS0.6.1 and lmfit1.3.4 metadata were checked, but no API examples or model fits for them were executed. They are optional alternatives, not dependencies of these scripts. See source ledger for evidence boundaries.

Raw lines 43 to 64