Inferring stochastic regulatory networks from perturbations of the non-equilibrium steady state

TL;DR

This paper introduces a novel method for inferring regulatory networks by analyzing fluctuations around the non-equilibrium steady state using a stochastic Gaussian mean-field approach, applicable to biological and neural systems.

Contribution

It develops a likelihood-based inference method leveraging stochastic modeling and Gaussian mean-field theory to improve network reconstruction from perturbation data.

Findings

Effective inference of regulatory interactions from perturbation data.

Application to both artificial and real phospho-proteomic datasets.

Comparison shows improved performance over steady-state-only methods.

Abstract

Regulatory networks describe the interactions between molecular or cellular regulators, like transcription factors and genes in gene regulatory networks, kinases and their receptors in signalling networks, or neurons in neural networks. A long-standing aim of quantitative biology is to reconstruct such networks on the basis of large-scale data. Our aim is to leverage fluctuations around the non-equilibrium steady state for network inference. To this end, we use a stochastic model of gene regulation or neural dynamics and solve it approximately within a Gaussian mean-field theory. We develop a likelihood estimate based on this stochastic theory to infer regulatory interactions from perturbation data on the network nodes. We apply this approach to artificial perturbation data as well as to phospho-proteomic data from cell-line experiments and compare our results to inference schemes restricted to mean activities in the steady state.