This is the code to accompany the paper "Sparse bayesian inference of mass-action biochemical reaction networks using the regularized horseshoe prior".
Examples of use and experiments presented in the paper are shown in the analysis directory in notebook format.
The RSindy abstract base class represents a class which
- Build an ansatz reaction set
- Construct the appropriate stocihiometric matrices, mass action reaction ratees, and descriptions
- Pre-process observational data
- Defines a method to estimate the reaction rate constants k as described in the paper
At the moment, we automatically construct the library of ansatz reactions as defined in utils.generate_valid_reaction_basis method.
The RSindyRegularizedHorseshoe implements the RSindy base class with the model and estimation techniques as defined in the paper.
Broadly, the RSindyRegularizedHorseshoe.fit_non_dx method generates a valid Stan model using the regularized horseshoe prior and the non-derivative observational model from the stored stoichiometric matrix and reaction rate vectors. Posterior distributions are estimated using a few default settings and the entire fit is directly returned to the user for further analysis as demonstrated in the analysis notebooks.