output
html_document pdf_document default default

A Neural-ODE approach for pharmacokinetics modeling and its advantage to alternative machine learning models in predicting new dosing regimens

The work is an application of Neural-ODE on the pharmacokinetics modeling tasks, and makes the comparisons with other machine learning models including LightGBM and LSTM, as well as the non-linear mixed-effects model (NLME). Please contact ([email protected] or [email protected]) if you have any questions or suggestions.

Workflow	Model Schemata

---

Required dependencies

Python main dependecies for Neural-ODE

• python (3.7.5)
• pytorch (1.5.1)
• torchdiffeq PyTorch implementation of differentiable ODE solvers

Install the dependencies with the provided environment file

• Use pip

pip install -r requirements.txt

Dataset

The clinical data used in this study is not made available due to reasons of patient privacy, but is available upon reasonable request from the authors with the approval of Genentech.

To train the models with the custom data, the table below lists the required columns:

Column name	Description
PTNM	patient number
STUD	study number
DSFQ	dosing frequency
CYCL	dosing cycles
AMT	dosing amounts
TIME	time in hours since the experiment begin for one individual
TFDS	time in hours since the last dosing
DV	the observations of PK

Models

Neural-ODE

All of the experiments in these models can directly run the run.sh in the following directories

• Neural-ODE 5-fold cross-validation

./5fold_models/Neural-ODE

• Neural-ODE cross dosing regimens

./cross-schedule_models/Neural-ODE
```# Neural_PK