NetDiff – Bayesian model selection for differential gene regulatory network inference

Thomas Thorne

DOI: https://doi.org/10.1038/srep39224

IF: 4.6

2016-12-01

Scientific Reports

Abstract:Differential networks allow us to better understand the changes in cellular processes that are exhibited in conditions of interest, identifying variations in gene regulation or protein interaction between, for example, cases and controls, or in response to external stimuli. Here we present a novel methodology for the inference of differential gene regulatory networks from gene expression microarray data. Specifically we apply a Bayesian model selection approach to compare models of conserved and varying network structure, and use Gaussian graphical models to represent the network structures. We apply a variational inference approach to the learning of Gaussian graphical models of gene regulatory networks, that enables us to perform Bayesian model selection that is significantly more computationally efficient than Markov Chain Monte Carlo approaches. Our method is demonstrated to be more robust than independent analysis of data from multiple conditions when applied to synthetic network data, generating fewer false positive predictions of differential edges. We demonstrate the utility of our approach on real world gene expression microarray data by applying it to existing data from amyotrophic lateral sclerosis cases with and without mutations in C9orf72, and controls, where we are able to identify differential network interactions for further investigation.

multidisciplinary sciences

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the change of gene regulatory network structures under different conditions (such as disease states and normal states, or responses to different stimuli). Specifically, the author proposes a new method named NetDiff for inferring differential gene regulatory networks from gene expression microarray data. This method is based on Bayesian model selection and identifies differential changes in gene regulatory networks by comparing conserved and changed network structure models. The main objective of the paper is to provide a more robust method to detect network edge changes under different conditions, reduce the probability of false predictions, and improve the understanding of cell process changes, especially when studying the differences in molecular mechanisms between cases and control groups. The core of the NetDiff method lies in applying variational inference techniques to learn Gaussian Graphical Models (GGM), which enables Bayesian model selection to be carried out with significantly higher computational efficiency than the Markov Chain Monte Carlo (MCMC) method. Through this method, the author was able to verify the effectiveness of their method on synthetic network data and real - world data sets, especially for the analysis of gene regulatory network differences between patients with Amyotrophic Lateral Sclerosis (ALS) and control groups, and between ALS patients with and without the C9orf72 gene mutation.

NetDiff – Bayesian model selection for differential gene regulatory network inference

Using Matrix of Thresholding Partial Correlation Coefficients to Infer Regulatory Network

A New Insight into Underlying Disease Mechanism Through Semi-Parametric Latent Differential Network Model

Differential network analysis by simultaneously considering changes in gene interactions and gene expression

A Bayesian model selection approach for identifying differentially expressed transcripts from RNA-Seq data

Bayesian variable selection and data integration for biological regulatory networks

Testing Differential Networks with Applications to Detecting Gene-by-Gene Interactions.

Approximate inference of gene regulatory network models from RNA-Seq time series data

Integrating gene regulatory pathways into differential network analysis of gene expression data

DifferentialRegulation: a Bayesian hierarchical approach to identify differentially regulated genes

Covariate-Adjusted Inference for Differential Analysis of High-Dimensional Networks

Bayesian Nonparametric Variable Selection as an Exploratory Tool for Finding Genes that Matter

Gene network reconstruction from transcriptional dynamics under kinetic model uncertainty: a case for the second derivative

WDNE: an integrative graphical model for inferring differential networks from multi-platform gene expression data with missing values

Differential analysis in Transcriptomic: The strength of randomly picking 'reference' genes

Differential network analysis from cross-platform gene expression data

Non-parametric Bayesian modelling of digital gene expression data

Identifying targets of multiple co-regulating transcription factors from expression time-series by Bayesian model comparison

Large-scale dynamic gene regulatory network inference combining differential equation models with local dynamic Bayesian network analysis

From Noise to Knowledge: Diffusion Probabilistic Model-Based Neural Inference of Gene Regulatory Networks

Network Module-Based Model in the Differential Expression Analysis for RNA-seq.