Finite time distributions of stochastically modeled chemical systems with absolute concentration robustness

TL;DR

This paper investigates the stochastic behavior of biochemical systems with absolute concentration robustness, showing that under certain scaling, their distributions converge to a Poisson distribution with mean equal to the deterministic ACR value.

Contribution

It proves that the distribution of ACR species in stochastic models converges to a product-form Poisson distribution, aligning stochastic and deterministic perspectives.

Findings

Distribution converges to Poisson with ACR mean

Supports conjectures on stochastic ACR behavior

Reconciles deterministic and stochastic model differences

Abstract

Recent research in both the experimental and mathematical communities has focused on biochemical interaction systems that satisfy an "absolute concentration robustness" (ACR) property. The ACR property was first discovered experimentally when, in a number of different systems, the concentrations of key system components at equilibrium were observed to be robust to the total concentration levels of the system. Followup mathematical work focused on deterministic models of biochemical systems and demonstrated how chemical reaction network theory can be utilized to explain this robustness. Later mathematical work focused on the behavior of this same class of reaction networks, though under the assumption that the dynamics were stochastic. Under the stochastic assumption, it was proven that the system will undergo an extinction event with a probability of one so long as the system is conservative, showing starkly different long-time behavior than in the deterministic setting. Here we consider a general class of stochastic models that intersects with the class of ACR systems studied previously. We consider a specific system scaling over compact time intervals and prove that in a limit of this scaling the distribution of the abundances of the ACR species converges to a certain product-form Poisson distribution whose mean is the ACR value of the deterministic model. This result is in agreement with recent conjectures pertaining to the behavior of ACR networks endowed with stochastic kinetics, and helps to resolve the conflicting theoretical results pertaining to deterministic and stochastic models in this setting.