ChemODE

Official implementation for the paper "ChemODE: A Physics-Informed Neural Surrogate for Robust Pharmacokinetics and Biochemical Dynamics".

Datasets

To ensure full statistical reproducibility, the raw synthetic time-series data used to benchmark the model are provided in this repository:

• synthetic_PK_data.csv: 3-compartment pharmacokinetic system with non-linear clearance.
• synthetic_Enzyme_data.csv: Isolated enzyme saturation dynamics (ablation study).
• synthetic_Brusselator_data.csv: Non-linear biological limit cycle oscillations.

Note: The real-world clinical Theophylline dataset is sourced from the open-source NONMEM repository as cited in the manuscript.

Usage

1. Open the notebook ChemODE_Revision.ipynb in Google Colab.
2. Set Runtime to T4 GPU.
3. Run all cells to reproduce Figures 1-3 and results from the manuscript.

Dependencies

• torchdiffeq
• pysindy
• torch
• numpy
• pandas