fluxtools sets up and solves flux balance analysis problems with nonlinear constraints. Utilities for creating large-scale models from copies of a base model are also included.

Create a network object by loading a metabolic model from sbml:

import fluxtools.sbml_interface as si
net = si.fromSBMLfile('example.xml')

or from a dictionary of reaction stoichiometries:

import fluxtools.reaction_networks as rn
stoichiometries = {'tx_A': {'A_ext': -1., 'A': 1.},
                   'R1': {'A': -1, 'B': 1},
                   'R2': {'A': -1, 'C': 1},
                   'R3': {'B': -1, 'D': 1},
                   'R4': {'C': -1, 'D': 1},
                   'sink': {'D': -1}}
net = rn.net_from_dict(stoichiometries)

Then, convert it to an optimization model:

from fluxtools.nlcm import NonlinearNetworkModel
model = NonlinearNetworkModel('example', net,
                              external_species_ids=('A_ext',))
model.set_objective('max_sink', '-1.0*sink')

Note the objective function is minimized. The basic FBA steady-state constraints are incorporated automatically (specified external species are not conserved.)

Add additional constraints (at least one nonlinear constraint must be given):

model.add_constraint('quadratic', 'R2-R1**2', 0.)
model.compile() 

Set upper and lower bounds on variables:

model.set_bound('R3', (0., 2.))

Solve, and interact with the solution:

>>> model.solve() # optionally, specify an initial guess as numpy array
[... IPOPT output omitted...]
array([ 2.00000002,  4.00000008,  2.        ,  4.00000008,  6.0000001 ,
        6.0000001 ])
>>> model.obj_value
-6.0
>>> model.soln['R2']
4.0

The model may be adjusted and re-solved, but model.compile() should be called every time a constraint is added or removed.

The package itself can simply be installed with python setup.py install once all dependencies are available.

• Numpy
• Scipy
• nose
• libsbml python bindings, http://sbml.org/Software/libSBML
• pyipopt (recommended version: http://github.com/ebogart/pyipopt)
• IPOPT, https://projects.coin-or.org/Ipopt
• pyglpk, http://tfinley.net/software/pyglpk/ (change '-m32' options in setup.py to '-m64' if necessary and be prepared to ignore some irrelevant failing tests with recent versions of glpk)
• glpk, https://www.gnu.org/software/glpk/

Tests of much of the package's functionality are included. Run 'nosetests' in the source directory to run the tests.

• nlcm - nonlinear optimization models with IPOPT
• stoichiometry_matrix - strictly linear FBA models with GLPK
• sbml_interface - read and write models in SBML format
• gene_utilities - load and save (some) GPR associations
• deblock - identify and purge blocked reactions and singleton metabolites based on the structure of the metabolic network
• fluxmap - trace which reactions consume/produce individual species in a solution
• c4clone - duplicate a network and connect the copies, as when setting up bundle sheath and mesophyll cells in C4 plants
• expr_manip - differentiation and parsing tools
• reaction_networks - provides classes representing models and their components
• simplification - simplify models by examining their linear subproblems
• functions - classes for representing constraint/objective functions
• keyedlist - the KeyedList data structure
• utilities - assorted tools for dealing with COBRA-compliant SBML files, total flux objectives, and analyzing optimization problem structure
• test - test suite

A (slightly) more detailed introduction for users and notes on numerical considerations are in preparation.

Eli Bogart, [email protected]

Code for the expr_manip, sbml_interface, keyedlist, and reaction_networks modules originated in SloppyCell, http://sloppycell.sourceforge.net/