NETSCOPE: Information-Theory based Network Discovery and Analysis

Abstract

BackgroundReconstruction of molecular networks allows structural and functional characterization of cells, tissues, and organs. A network analytical approach to statistical correlations in molecular data can provide new insight into the underlying system, but currently there is not any analytical approach that is generalizable across different types of data. Here we present TITAN, an open-source toolbox in MATLAB and Octave for the reconstruction and graph analysis of molecular networks. Using an information-theoretical approach TITAN reconstructs networks from transcriptional data, revealing the topological structure of correlations in biological systems.FindingsThe toolbox includes a data processing pipeline to construct gene co-expression networks from transcriptional data, mutual information (MI) and variation of information (VI) to quantify connectivity and graph-bas ed analysis for molecular network analysis. We show that the toolbox can accurately reconstruct network structure from synthetic data with predetermined network topology. We use the toolbox to construct the gene co-expression network of S. cerevisiae and show that the resulting network is indistinguishable from gene co-expression networks previously identified experimentally.DiscussionTITAN is a general tool for network reconstruction in high dimensional datasets. It could be used to study co-transcriptional networks and reveal connectivity patterns in any large omics datasets. Considering the universal nature of MI as a measure of correlation, TITAN can potentially be used to vertically integrate datasets for a cross-omics analysis that covers a system on every level, from genes to metabolites. This universality can be extended to cross-species analysis and evolutionary genetics. Other applications include molecular fingerprinting of cells, tissues, and organs, and tracing developmental trajectories for gene and protein networks.