Polynomial representations of piecewise-linear differential equations arising from gene regulatory networks (Q1940179)

The paper investigates a system of ODEs of the form \[ dx_i/dt=F_i(z_1,z_2,\dotsc ,z_n)-G_i(z_1,z_2,\dotsc ,z_n)x_i,\;i=1,2,\dotsc ,n, \] modelling the dynamics of gene product concentrations \(x_i(t)\) in a gene regulatory network. Here \(z_k=z_k(x_k)\) are the response functions which are assumed to be step functions representing whether the corresponding gene \(k\) is activated or deactivated provided that the product concentration \(x_k\) is below or above a certain threshold, say \(\theta_k\). Finally, \(F_i\) and \(G_i\) are the so called regulatory functions which depend on the response functions \(z_k\). The study is based on singular perturbation analysis where the step functions are approximated by (smooth) Hill functions: \(z_k=H(x_k,\theta_k,q_k)=x_k^{1/q_k}/\left(x_k^{1/q_k}+\theta_k^{1/q_k}\right)\). It is shown that, inside any regular domain, i.e., a domain of the phase space such that no threshold is crossed, the limit dynamics as \(q_k\downarrow0\) could be completely described by a uniquely defined equivalent system with multilinear regulatory functions replacing the original. On the other hand, the limit dynamics in the singular domains is sensitive to the particular form of the functions \(F_i\) and \(G_i\), and their replacement with multilinear functions may not be accurate. Nevertheless, the authors propose a solution of that problem substituting the regulatory functions with higher-order polynomial representations. Finally, for that purpose, the minimal polynomial degree is determined in the generic cases.