AI-PBPK-MPC

Research-grade, end-to-end pipeline for AI-parameterized PBPK simulation with MPC and reporting.

Quickstart

1. Install Poetry and dependencies

poetry install

2. Drop your Nano-Tumor Excel/CSV(s) into data/raw/.

3. Run an experiment (E2 canonical case):

python scripts/run_experiment.py --exp e2 --config ai_pbpk/config/experiments/e2.yaml

Artifacts are saved under artifacts/<timestamp>/. Figures under reports/figures/ and reports/summary.pdf via:

python -m ai_pbpk.report.build_report

Make targets

make setup   # poetry install
make test    # pytest -q
make run-e2  # run canonical E2

IPOPT/CasADi install notes

• CasADi Python wheels are installed via Poetry (casadi>=3.6).
• IPOPT is used by CasADi as a native solver; on Windows/Linux, install a prebuilt IPOPT binary and ensure it is on PATH. See https://coin-or.github.io/Ipopt/INSTALL.html for details. If IPOPT is unavailable, the MPC uses a SciPy SLSQP fallback.

Project layout

See ai_pbpk/ for modules: data prep, AI parameter predictor, PBPK ODE simulator, MPC, experiments, evaluation, visualization, and report.

Reproducibility

• All runs log seeds, config, and git commit to MLflow (local mlruns/).
• Configurable via YAML in ai_pbpk/config/.