PKPDBuilder

AI/ML Pharmacokinetic Modeling

Upload your PK data and compare traditional compartmental models against modern AI/ML approaches — Neural ODEs, XGBoost, and Gaussian Process regression — all in one analysis. See which approach best captures your drug's pharmacokinetics.

Traditional Compartmental

1-CMT and 2-CMT models with first-order absorption. The gold standard.

scipy.optimize

Neural ODE

A neural network learns the ODE system governing PK dynamics directly from data.

PyTorch + torchdiffeq

ML Ensemble

XGBoost for feature-driven prediction + Gaussian Process for uncertainty quantification.

XGBoost + scikit-learn

Upload PK Data

📄
Drop CSV file here or click to browse
Required columns: time, conc (or dv/cp). Optional: dose, ID
Or try sample data:

How It Works

1

Upload Your Data

Upload a CSV with time and concentration columns. Additional columns (dose, subject ID, covariates) improve the analysis. Standard NONMEM-style datasets work directly.

2

Automated Model Fitting

We fit 4 models simultaneously: 1-compartment, 2-compartment, Neural ODE (learning PK dynamics from scratch), and an XGBoost + Gaussian Process ensemble.

3

Compare & Interpret

Side-by-side comparison of all approaches with RMSE, R², AIC, diagnostic plots, and model-specific insights. Understand when AI/ML adds value for your data.

Expected CSV Format

ID,time,conc,dose
1,0.5,2.341,500
1,1.0,5.672,500
1,2.0,8.123,500
...

Column names are flexible — we auto-detect time (t, time, hours), concentration (conc, dv, cp, c), dose (amt, dose), and subject ID (id, subject). Case-insensitive.

Python Packages Used

PyTorch v2.10
Neural network framework
torchdiffeq v0.2.5
Neural ODE solver (dopri5)
XGBoost v3.2
Gradient boosted trees
scikit-learn v1.8
Gaussian Process regression
SciPy v1.17
Traditional model optimization
lmfit v1.3
Parameter estimation
PKPDBuilder.com · For research and educational purposes only. Not for clinical decision-making.