Reverse-engineering transcriptional modules from gene expression data

TL;DR

This paper introduces a probabilistic framework for reverse-engineering gene regulatory modules from expression data, enabling the inference of robust, significant modules and their regulators that generalize beyond the training data.

Contribution

The authors develop a novel method to infer ensembles of module networks and extract the most significant modules and regulators through an averaging procedure.

Findings

Inferred models generalize beyond the training data

Method identifies statistically significant modules

Enables robust gene regulatory network inference

Abstract

"Module networks" are a framework to learn gene regulatory networks from expression data using a probabilistic model in which coregulated genes share the same parameters and conditional distributions. We present a method to infer ensembles of such networks and an averaging procedure to extract the statistically most significant modules and their regulators. We show that the inferred probabilistic models extend beyond the data set used to learn the models.