ibuffpy 0.1.0

A Python package for appling Structural Sensitivity Analysis to reaction networks.

ibuffpy

This package applies structural sensitivity analysis to chemical reaction networks to discover buffering structures within them.

Installation

To install ibuffpy, run the following command:

pip install ibuffpy

The pygraphviz package is required to visualize the hierarchy graph of buffering structures. See https://pygraphviz.github.io/documentation/stable/install.html for the installation of pygraphviz.

Usage

Once you have installed ibuffpy, you can import ReactionNetwork module from the ibuffpy package

from ibuffpy import ReactionNetwork

In ibuffpy, reaction networks are treated as instances of the ReactionNetwork.ReactionNetwork which contains information about reaction names, metabolites, and stoichiometric matrices. To read a network from a csv file, use ReactionNetwork.from_csv(" PATH_TO_CSV") where the csv file must contain reaction formulas such as "reaction_idx, substrate1 substrate2, product1 product2" (an example can be found in demo_network1.csv). If the reaction equation contains more than one metabolite, separate them with a space. If the coefficient of a metabolite is $n$, it will appear n times on one side. Outflow reactions or inflow reactions are represented as ['inflow_name',['metabolite_name'],['out']] or ['outflow_name',['out'],['metabolite_name']]. The metabolite name 'out' is reserved for the node corresponding to outside of the network, which does not appear in the stoichiometric matrix.

network = ReactionNetwork.from_csv("PATH_TO_CSV")
network.info()
# list of reactions in the network
print(network.reaction_list)
# list of metabolites in the network except for 'out' node
print(network.cpd_list_nooout)

Structural sensitivity analysis computes the network sensitivity to parameter perturbation from the A-matrix, which reflects network structure. ReactionNetwork.compute_amat(network) returns an A-matrix with random real values in nonzero entries of $\partial \boldsymbol r / \partial \boldsymbol x$, a basis of $\ker \nu$, and a basis of $\ker \nu^\top$. ReactionNetwork class uses scipy.linalg.null_space to obtain bases by default, but you can also use ker_basis='rref' option. ker_bais or coker_basis accepts ndarray or list to provide a self-made bases.

# numpy is used to calculate a basis by default.
network = ReactionNetwork.from_csv(path_to_csv, ker_basis='svd', coker_basis='rref')

# augumented matrix
amat = network.compute_amat()

# sensitivity matrix
smat = np.linalg.inv(amat)

You can find buffering structures by

bs_list = ReactionNetwork.compute_bs(network)

and visualise them by

graph = ReactionNetwork.make_hiergraph(bs_list)
graph.draw(PATH_SAVE)