Identifiability of linear noise approximation models of chemical reaction networks from stationary distributions

Abstract

Biomolecular systems can often be modeled by chemical reaction networks with unknown parameters. In many cases, the available data is constituted of samples from the stationary distribution, wherein each sample is given by a cell in a population. In this work, we develop a framework to assess identifiability of parameters in such a situation. Working with the Linear Noise Approximation (LNA) we give an algebraic formulation of identifiability and use it to certify identifiability with Hilbert’s Nullstellensatz. We include applications to particular biomolecular systems, focusing on the identifiability of a sequestration-based motif and of a feedback arrangement based on it.