How to infer gene networks from expression profiles, revisited

Abstract

Inferring the topology of a gene-regulatory network (GRN) from genome-scale time-series measurements of transcriptional change has proved useful for disentangling complex biological processes. To address the challenges associated with this inference, a number of competing approaches have previously been used, including examples from information theory, Bayesian and dynamic Bayesian networks (DBNs), and ordinary differential equation (ODE) or stochastic differential equation. The performance of these competing approaches have previously been assessed using a variety of in silico and in vivo datasets. Here, we revisit this work by assessing the performance of more recent network inference algorithms, including a novel non-parametric learning approach based upon nonlinear dynamical systems. For larger GRNs, containing hundreds of genes, these non-parametric approaches more accurately infer network structures than do traditional approaches, but at significant computational cost. For smaller systems, DBNs are competitive with the non-parametric approaches with respect to computational time and accuracy, and both of these approaches appear to be more accurate than Granger causality-based methods and those using simple ODEs models.

Mathematical Modelling Plant Signalling Networks

“…The performance of a continuous, linear-Gaussian state-space model proposed in [30] has been tested on a well-characterised Arabidopsis clock network using data from replicated time course microarrays [66]. The method was able to recover known links accurately, including the feedback loop between morning and evening elements of the clock, as well as to capture partially loops between some clock-related genes.…”

Arabidopsis Defense against Botrytis cinerea: Chronology and Regulation Deciphered by High-Resolution Temporal Transcriptomic Analysis

“…The discrete-time causal structure identification (CSI) algorithm of Klemm (Klemm, 2008;Penfold and Wild, 2011) was used to infer a regulatory network from the cluster means and B. cinerea growth. A section of the predicted network is shown in Figure 11, and several regulatory predictions can be made from this network.…”

“…The mean of each cluster was taken as being representative of that particular cluster and a network inferred using CSI (Klemm, 2008;Penfold and Wild, 2011). An additional node in the network representing B. cinerea tubulin expression during infection was included.…”

CSI: a nonparametric Bayesian approach to network inference from multiple perturbed time series gene expression data

“…The posterior distribution over parental sets (and hyperparameters) can be constructed via Bayes' rule, and consists of combinatorially searching through all sets of putative regulators (up to a maximum in-degree d), and assigning a likelihood to each set. We may obtain a point estimate of the distribution via an Expectation Maximisation algorithm (Penfold and Wild, 2011), or sample from it via MCMC (Penfold et al, 2012). The Gaussian process model underpinning the dynamics of gene regulation requires a matrix inversion (which scales as O(m 3 ) where m is the number of experimental observations) for each possible parental set and each step in the gradient optimisation.…”

“…Whilst most methods for inferring GRNs from time series mRNA expression data are only able to cope with single time series (or single perturbations with biological replicates), it is becoming increasingly common for several time series to be generated under different experimental conditions. The CSI algorithm (Klemm, 2008) represents one approach to inferring GRNs from multiple time series data, which has previously been shown to perform well on a variety of datasets (Penfold and Wild, 2011). Another challenge in network inference is the identification of condition specific GRNs i.e., identifying how a GRN is rewired under different conditions or different individuals.…”