[Paper Review] Dynamical Equivalence and Linear Conjugacy of Chemical Reaction Networks: New Results and Methods

Matthew D. Johnston, David Siegel+1 more|arXiv (Cornell University)|Nov 6, 2011

Computational Drug Discovery Methods29 references13 citations

TL;DR

This paper introduces mixed-integer linear programming (MILP) methods to compute dense, sparse, reversible, detailed balanced, and complex balanced linearly conjugate reaction networks under mass-action kinetics. It proves that dense linearly conjugate networks are uniquely determined by the complex set and that equilibrium choice does not affect the existence of complex or detailed balanced realizations, enabling structure-based analysis of reaction mechanisms without prior rate constants.

ABSTRACT

In the first part of this paper, we propose new optimization-based methods for the computation of preferred (dense, sparse, reversible, detailed and complex balanced) linearly conjugate reaction network structures with mass action dynamics. The developed methods are extensions of previously published results on dynamically equivalent reaction networks and are based on mixed-integer linear programming. As related theoretical contributions we show that (i) dense linearly conjugate networks define a unique super-structure for any positive diagonal state transformation if the set of chemical complexes is given, and (ii) the existence of linearly conjugate detailed balanced and complex balanced networks do not depend on the selection of equilibrium points. In the second part of the paper it is shown that determining dynamically equivalent realizations to a network that is structurally fixed but parametrically not can also be written and solved as a mixed-integer linear programming problem. Several examples illustrate the presented computation methods.

Motivation & Objective

• To develop optimization-based methods for computing preferred linearly conjugate reaction network structures with specific dynamical properties such as reversibility, detailed balance, and complex balance.
• To extend existing MILP frameworks to handle networks with minimal and maximal numbers of complexes, reactions, and complexes under structural constraints.
• To address the problem of dynamical equivalence when network structure is fixed but rate constants are unknown, enabling structure-only analysis of reaction mechanisms.
• To prove theoretical results on the uniqueness of dense linearly conjugate networks and the invariance of complex/detailed balancing existence across different equilibrium points.
• To broaden the applicability of reaction network theory by enabling parameter-free structural analysis of dynamical equivalence and conjugacy.

Proposed method

• Formulate the search for linearly conjugate networks as a mixed-integer linear programming (MILP) problem, leveraging sparsity and structural constraints.
• Use stoichiometric and complex matrix representations to encode network structure and dynamics, with rate constants as variables in the optimization.
• Apply constraints to enforce weak reversibility, reversibility, detailed balance, and complex balance via linear equality and inequality conditions.
• Introduce a transformation-based approach to map dynamics between networks via non-trivial linear conjugacy, preserving qualitative behavior.
• Utilize positive diagonal state transformations to relate dynamical equivalence and conjugacy, enabling equilibrium-invariant analysis.
• Extend the MILP framework to cases where rate constants are not fixed, allowing structural analysis independent of specific kinetic parameters.

Experimental results

Research questions

• RQ1 Can we compute the sparsest and densest linearly conjugate networks with specified dynamical properties (e.g., reversibility, detailed balance) using optimization?
• RQ2 Is the structure of the densest linearly conjugate network uniquely determined by the set of complexes and a positive diagonal transformation?
• RQ3 Does the existence of a linearly conjugate complex balanced or detailed balanced network depend on the choice of equilibrium point?
• RQ4 Can dynamical equivalence be determined and computed when only the network structure is known, not the rate constants?
• RQ5 How can we systematically generate alternative realizations of a reaction network that preserve dynamics under mass-action kinetics but vary in reaction structure?

Key findings

• Dense linearly conjugate networks are uniquely determined by the set of complexes and a positive diagonal state transformation, regardless of rate constants.
• The existence of linearly conjugate detailed balanced and complex balanced networks is independent of the choice of equilibrium point, enabling flexible construction from any positive equilibrium.
• The unweighted directed reaction graph of any linearly conjugate network is a subgraph of the densest such network, ensuring structural hierarchy.
• The problem of finding sparse and dense realizations under weak reversibility, reversibility, and complex/detailed balancing can be formulated as a MILP, enabling efficient computation.
• When rate constants are not fixed, the computation of dynamically equivalent realizations with minimal or maximal number of complexes remains a MILP problem, allowing structure-only analysis.
• Complex balanced networks with the same structure as dynamically equivalent realizations can be constructed by reassigning rate constants, demonstrating parameter flexibility in network design.