Reconstruction of gene regulatory networks from steady state data

TL;DR

This paper introduces a novel method to reconstruct gene regulatory network Jacobian matrices from steady state protein data, enabling better understanding of gene interactions and network properties.

Contribution

The authors develop a new approach to estimate the Jacobian matrix of gene regulatory systems using equilibrium data, linking observable functions to unobservable network connectivity.

Findings

High accuracy in Jacobian estimation with noiseless data (up to 100%)

Effective application to Drosophila and simulated gene networks

Provides sign and magnitude estimates of network interactions

Abstract

Genes are connected in regulatory networks, often modelled by ordinary differential equations. Changes in expression of a gene propagate to other genes along paths in the network. At a stable state, the system's Jacobian matrix confers information about network connectivity. To disclose the functional properties of genes, knowledge of network connections is essential. We present a new method to reconstruct the Jacobian matrix of models for gene regulatory systems from equilibrium protein concentrations. In a recent paper we defined propagation and feedback functions describing how genetic variation at one gene propagates to the other genes in the network and possibly also back to itself. Here we show how propagation and feedback functions provide relations between equilibrium protein levels which are in principle observable, and Jacobi elements which are not directly observable. We establish exact formulae expressing the Jacobian in terms of derivatives of propagation and feedback functions. Approximating these derivatives from perturbed and unperturbed protein levels, we derive formulae for estimating the Jacobian. We apply the method to models of the Drosophila segment polarity network and randomly generated gene networks. Genes could be perturbed in two ways: by modifying mRNA degradation rates, or by allele knockout in diploid models. Comparison with the true Jacobians shows that for noiseless data we obtain hit rates close to 100% in the former case and in the range 80-90% in the latter. Our method adds to the network interference toolbox and provides a sign estimate of the Jacobian from steady state data, and a value estimate of the Jacobian if protein degradation rates are known. Also the approach identifies some predicted connections as much more reliable than others, and could point to further experiments for resolving uncertainties in the less dependable Jacobian elements.