PK/PD Modelling & Simulation

What it does

Turns Claude Code into a pharmacometrics assistant backed by nine Python scripts, each answering one question:

• nca.py — non-compartmental exposure metrics (AUC, Cmax, lambda_z, t½, CL/F) with explicit AUC method, BLQ rule and lambda_z window; raises findings for >20% extrapolation or a terminal phase that was never reached
• fit_compartmental.py — 1/2/3-compartment comparison with AIC, BIC and F-test, log-scale estimation, RSE and parameter-correlation checks, plus a runs test that separates structural misspecification from a wrong error model
• check_popk_dataset.py — catches the silent NONMEM dataset defects (BLQ text read as zero, blank covariates becoming 0 kg, ADDL without II, duplicate timestamps)
• simulate_regimen.py — steady-state and population simulation with target-attainment fractions
• exposure_response.py — Emax/sigmoid ER plus ICH E14-style C-QTc using the 90% CI upper bound
• bioequivalence.py — ABE vs EMA ABEL vs FDA RSABE, with sample size and exact power
• allometry_and_fih.py — allometric scaling with Anderson-Holford maturation, MRSD and MABEL
• ddi_static.py — ICH M12 basic and mechanistic static models with cut-offs
• tdm_bayes.py — MAP Bayesian individual parameter estimation from measured levels

Every script supports --format table|tsv|json, sends data to stdout and findings to stderr, and exits 0/1/2 so it can gate a pipeline. Fourteen reference documents cover NCA conventions, popPK, PBPK, special populations, regulatory guidance and the software ecosystem.

Who it's for

• Pharmacokineticists and pharmacometricians in pharma or CROs
• Anyone preparing datasets and model assumptions before NONMEM, Monolix or nlmixr2
• Regulatory scientists who need the right bioequivalence criterion applied to the right design
• Graduate students who want the reasoning, not just the numbers

Examples

1. NCA reporting — run python3 nca.py -i profile.csv --dose 100 --route extravascular --partial-auc 0-24; the finding that 25% of AUCinf is extrapolated justifies reporting AUC(0-tau) at steady state instead.
2. Model selection defence — AIC prefers a 3-compartment model, but Q3 has 98% RSE; use the BIC, F-test and RSE output to document why the 2-compartment model was retained.
3. Regimen choice — simulate 500 mg q12h and show only 44% of the population attains a trough of 4 mg/L, then compare alternative regimens on attainment rather than the typical patient.

Scope note: the scripts compute and diagnose; they never conclude. Bioequivalence decisions, patient dosing and QT liability remain with qualified clinicians and regulators, and the bundled vancomycin model is explicitly illustrative.

Install the skill I found on Claude Skill Mart.
Copy the skills/pkpd-modeling folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/pkpd-modeling/.
When it's done, tell me in one line what this skill can do.