OpenPKFlow

What it does

Install

pip install openpkflow

pip install openpkflow[reports]

pip install openpkflow[bayes]

Quick start: dissolution similarity

from openpkflow.dissolution import f1, f2
reference = [20.0, 40.0, 60.0, 80.0, 90.0]
test = [21.0, 39.0, 61.0, 79.0, 88.0]
print(f"f1 = {f1(reference, test):.2f}")
print(f"f2 = {f2(reference, test):.2f}")

From a CSV file

from openpkflow.dissolution import DissolutionStudy
study = DissolutionStudy.from_csv("dissolution.csv")
# or load directly from Excel (requires pip install openpkflow[reports]):
# study = DissolutionStudy.from_excel("dissolution.xlsx", sheet_name="Data")
result = study.compare(reference="reference", test="test")
result.summary()
result.report("dissolution_report.html")
result.report("dissolution_report.pdf", format="pdf") # requires [reports]

Advanced dissolution workbench

from openpkflow.dissolution import (
 DissolutionWorkbenchConfig,
 run_dissolution_workbench_csv,
)
config = DissolutionWorkbenchConfig(
 reference_label="reference",
 test_label="test",
 bootstrap_replicates=5000,
 seed=2026,
)
workbench = run_dissolution_workbench_csv("vessel_profiles.csv", config)
print(workbench.comparison.summary())
print(workbench.reference_models.summary())
workbench.report("dissolution_workbench.html")
workbench.audit_bundle("dissolution_workbench_audit.zip")

CLI

openpkflow version
openpkflow similarity --reference "20,40,60,80" --test "21,39,61,79"

Quick start: NCA

from openpkflow.nca import NCAStudy
study = NCAStudy.from_csv(
 "pk_data.csv",
 auc_method="linear_up_log_down", # required: "linear", "log", or "linear_up_log_down"
 blq_method="none", # required: "none", "drop", "zero", "half_lloq", "lloq"
)
summary = study.analyze()
print(summary.summary()) # tabular ASCII output
# Per-subject results
result = summary.results[0]
print(f"Subject: {result.subject}")
print(f"AUClast: {result.AUClast:.2f} h*mg/L")
print(f"Cmax: {result.Cmax:.2f} mg/L")
print(f"Tmax: {result.Tmax:.2f} h")
print(f"t1/2: {result.half_life:.2f} h")
print(f"CL/F: {result.CL_F:.2f} L/h")
# Reports
result.report("nca_subject1.html")
summary.report("nca_summary.html")

NCA CSV format

subject,time,conc,dose,route
1,0.0,0.0,320.0,oral
1,0.5,4.2,320.0,oral
1,1.0,8.1,320.0,oral
1,2.0,6.8,320.0,oral
1,4.0,3.5,320.0,oral
1,8.0,1.7,320.0,oral
1,12.0,0.9,320.0,oral
1,24.0,0.2,320.0,oral

Quick start: PK simulation

import numpy as np
from openpkflow.sim import simulate
from openpkflow.sim.models import OneCompartmentModel
from openpkflow.sim.dosing import DoseRegimen
model = OneCompartmentModel(route="oral", CL_F=5.0, Vz_F=50.0, ka=1.2)
regimen = DoseRegimen.from_repeated(amount=100.0, route="oral", tau=24.0, n_doses=3)
times = np.linspace(0, 72, 500)
result = simulate(model, regimen, times)
print(result.summary())
result.report("sim_report.html")
result.report("sim_report.pdf", format="pdf") # requires [reports]

Quick start: Bayesian individual PK (MAP)

from openpkflow.bayes import map_individual_pk, PKPrior
import math
# Noiseless 1-cmt oral data (CL_F=5, Vz_F=50, ka=1.2, dose=100)
times = [0.5, 1.0, 2.0, 4.0, 8.0, 12.0]
concs = [1.23, 1.85, 1.97, 1.61, 0.89, 0.49]
result = map_individual_pk(times, concs, dose=100.0, route="oral", subject="S01")
print(result.summary()) # MAP estimates, SEs, diagnostics, disclaimer
result.report("map_pk_report.html")

from openpkflow.bayes.bayes_pk import bayes_individual_pk
result = bayes_individual_pk(times, concs, dose=100.0, route="oral",
 n_samples=1000, tune=1000, chains=2)
print(f"CL_F = {result.cl_mean:.3g} [95% CrI: {result.cl_95ci[0]:.3g}, {result.cl_95ci[1]:.3g}]")
print(f"P(shrinkage) = {result.shrinkage_cl:.1%}")

Quick start: Bayesian bioequivalence (requires [bayes])

import pandas as pd
from openpkflow.bayes.bayes_be import bayes_be
# Long-format 2x2 crossover data
data = pd.DataFrame({
 "subject": ["S01","S01","S02","S02","S03","S03","S04","S04"],
 "sequence": ["RT", "RT", "TR", "TR", "RT", "RT", "TR", "TR"],
 "period": [1, 2, 1, 2, 1, 2, 1, 2 ],
 "treatment": ["R", "T", "T", "R", "R", "T", "T", "R" ],
 "value": [98.0, 103.0, 95.0, 91.0, 107.0, 112.0, 99.0, 94.0],
})
result = bayes_be(data, metric="AUC", n_samples=2000, tune=1000, chains=2)
print(f"P(BE) = {result.p_be:.3f}")
print(f"GMR = {result.gmr_mean:.4g} [95% CrI: {result.gmr_95ci[0]:.4g}, {result.gmr_95ci[1]:.4g}]")
print(f"Frequentist 90% CI: [{result.freq_90ci[0]:.4g}, {result.freq_90ci[1]:.4g}]")
result.report("bayes_be_report.html")

Quick start: bioequivalence

import pandas as pd
from openpkflow.be import BEStudy
# Wide-format DataFrame: one row per subject, reference and test PK parameter values
be_df = pd.DataFrame({
 "subject": ["S01", "S02", "S03", "S04", "S05", "S06"],
 "sequence": ["RT", "RT", "RT", "TR", "TR", "TR"],
 "reference": [100.2, 98.7, 105.1, 97.3, 102.8, 99.5],
 "test": [95.1, 94.0, 99.8, 92.9, 97.4, 94.8],
})
study = BEStudy(be_df, parameter="AUCinf")
result = study.analyze() # default: 80-125%, alpha=0.05
print(result.summary())
result.report("be_report.html")
# NTI products: pass narrower limits
result_nti = study.analyze(be_lower=0.90, be_upper=1.1111)

From NCAStudy results (convenience)

from openpkflow.be import BEStudy
# Run NCA separately on each formulation's PK data
# reference_nca_summary = NCAStudy.from_csv("ref_pk.csv", ...).analyze()
# test_nca_summary = NCAStudy.from_csv("test_pk.csv", ...).analyze()
study = BEStudy.from_nca_results(
 reference_nca_summary, test_nca_summary, parameter="AUCinf"
)
result = study.analyze()

Formal complete balanced 2x2 BE ANOVA

from openpkflow.be import formal_be_anova
formal_result = formal_be_anova(long_be_df, parameter="AUCinf")
formal_result.report("formal_be_report.html")

from openpkflow.be import fda_partial_replicate_rsabe
rsabe_result = fda_partial_replicate_rsabe(partial_replicate_df, parameter="AUC")
rsabe_result.report("rsabe_report.html")

CLI

openpkflow be compare be_data.csv --parameter AUCinf --report be_report.html
openpkflow be anova formal_be_data.csv --parameter AUCinf --report formal_be_report.html

Quick start: population PK diagnostics

import pandas as pd
from openpkflow.pop import GOFResult, simulate_vpc
from openpkflow.sim.models import OneCompartmentModel
from openpkflow.sim.dosing import DoseRegimen
# GOF: supply your own PRED/IPRED from NONMEM or nlmixr2
gof = GOFResult(
 dv=[5.2, 8.1, 6.4, 3.2],
 pred=[4.9, 7.8, 6.0, 3.0],
 ipred=[5.1, 8.0, 6.3, 3.1],
 time=[1.0, 2.0, 4.0, 8.0],
 id=["S1", "S1", "S1", "S1"],
 sigma=0.15,
 study_label="Phase 1 Study",
)
print(gof.summary())
gof.report("gof_report.html")
# Simulation-based VPC
model = OneCompartmentModel(route="oral", CL_F=5.0, Vz_F=50.0, ka=1.2)
regimen = DoseRegimen.from_repeated(amount=100.0, route="oral", tau=24.0, n_doses=1)
observed = pd.DataFrame({"TIME": [1, 2, 4, 8, 12], "DV": [5.1, 8.2, 6.5, 3.8, 2.1]})
vpc = simulate_vpc(model, regimen, observed, n_replicates=500, seed=42)
vpc.report("vpc_report.html")

Feature comparison

Roadmap

Current status

By the numbers

Validation

Disclaimer

Documentation

Contributing

Citation

Thakar, P. (2026). OpenPKFlow: Python-first pharmacometrics and dissolution toolkit.
https://github.com/priyamthakar/openpkflow

License