The initial effort will focus on deriving chemical rate equations from concentration time-series data based on the law of mass action, i.e. systems of first-order ODEs with only polynomial terms on the right-hand side.
The following tools/principles are used.
-
Neural ODE by Ricky T.Q. Chen et al. (https://github.com/rtqichen/torchdiffeq)
-
Symbolic regression
-
Sparse regression
-
Knowledge of kinetic differential equations
After download the code, you can do the following
python3 -m venv pon-env
source pon-env/bin/activate
python setup.py develop
This should set 'train_poly' in your $PATH to use.