PyMixEF

PyMixEF is a Python-native platform for mixed-effects statistics and pharmacometrics. Its architecture follows one central rule: define the scientific model once in a typed, versioned intermediate representation, then let compatible estimation engines consume that representation.

Documentation · PyPI · Source · Issue tracker

Install

PyMixEF requires Python 3.11 or newer.

Published releases install from PyPI with:

python -m pip install pymixef

For development from a source checkout:

python -m pip install -e .

Tabular adapters, reports, and validation comparison dependencies are optional:

python -m pip install -e ".[data,report,validation]"

To run the tutorial notebooks from GitHub or the source archive, install their Jupyter runtime:

python -m pip install "pymixef[notebooks]"

The ten worked tutorials are backed by pre-executed Jupyter notebooks with reviewed results, multiple scientific figures, and assertion-backed checks. GitHub therefore renders complete worked examples before a reader launches a kernel.

Documentation

The complete documentation includes:

• installation profiles and a five-minute quickstart;
• an analysis chooser and full LMM, GLMM, MMRM, PK, ODE, and pharmacometrics guides;
• data, formula, family, link, covariance, inference, diagnostic, simulation, provenance, validation, CLI, and interoperability reference material;
• deep dives for all ten pre-executed notebooks with all 31 validated plots and exact saved results;
• a task-oriented map plus generated signature reference for every public Python module;
• a searchable catalog of all 56 evidence-gated capabilities.

To build and preview the documentation locally:

python -m pip install -e ".[docs]"
make docs
python -m http.server 8765 --bind 127.0.0.1 --directory docs/_build/dirhtml

The strict build fails on Sphinx warnings, stale notebook figures, missing API docstrings/search entries, broken tutorial coverage, or figure-manifest drift.

First linear mixed model

import pymixef

data = {
    "change": [2.1, 3.2, 2.8, 4.4, 3.7, 5.1],
    "time": [0, 1, 2, 0, 1, 2],
    "subject": ["A", "A", "A", "B", "B", "B"],
}

result = pymixef.fit(
    "change ~ time + (1 | subject)",
    data=data,
    method="reml",
)

print(result.summary())
result.save("fit.json")

Compile and inspect a model without fitting:

model = pymixef.Model.from_formula(
    "change ~ time + (1 | subject)",
    family=pymixef.families.Gaussian(),
)
plan = model.compile(data, engine="lmm", method="reml")
print(plan.explain())

Pharmacometric events and ODEs

from pymixef.pharmacometrics import canonicalize_events, simulate_ode

events = canonicalize_events(
    {
        "ID": [1, 1, 1],
        "TIME": [0.0, 1.0, 4.0],
        "EVID": [1, 0, 0],
        "AMT": [100.0, 0.0, 0.0],
        "CMT": [1, 1, 1],
    }
)

See the documentation map, the validation policy, the public validation artifacts, and the warning catalog.

Scientific and regulatory scope

Every likelihood path states its normalization and parameterization. Every fit retains a convergence object, run manifest, approximation method, data/model fingerprints, warnings, and serialized diagnostic inputs.

PyMixEF can support an organization's context-specific validation process. No package version is universally “FDA validated,” and this project makes no such claim.

License

Apache License 2.0. Formal dependency and name review remains the responsibility of distributors and adopters.