New methods for joint analysis of biological networks and expression data

Abstract

Biological networks, such as protein interaction, regulatory or metabolic networks, derived from public databases, biological experiments or text mining can be useful for the analysis of high-throughput experimental data. We present two algorithms embedded in the ToPNet application that show promising performance in analyzing expression data in the context of such networks. First, the Significant Area Search algorithm detects subnetworks consisting of significantly regulated genes. These subnetworks often provide hints on which biological processes are affected in the measured conditions. Second, Pathway Queries allow detection of networks including molecules that are not necessarily significantly regulated, such as transcription factors or signaling proteins. Moreover, using these queries, the user can formulate biological hypotheses and check their validity with respect to experimental data. All resulting networks and pathways can be explored further using the interactive analysis tools provided by ToPNet program.

Transforming omics data into context: Bioinformatics on genomics and proteomics raw data

“…Mathematical frameworks being applied in this context include Boolean networks [179][180][181], graphical Gaussian networks [182,183], Bayesian models [184,185], correlation analysis [186], graph theory [187,188], and genetic algorithms [189]. Software has been developed for these procedures as GeneTS, utilizing a graphical Gaussian network framework to compute a graph representing correlations between objects based on time series data.…”

Using Biological Networks in Protein Function Prediction and Gene Expression Analysis

“…The second group are the direct group analysis methods [23,38,79], which test whether a biological pathway is differentially expressed as a whole. The third group are the network-based analysis methods [15,73,77], which zoom into a subnetwork of a biological pathway and test whether the subnetwork is differentially expressed. All of these approaches have their basis on the fact that every disease phenotype has some underlying biological causes.…”

“…Then, in Section 4, the three groups of approaches [4,6,9,13,17,28,40,55,69,70] for prediction of protein function using biological networks are presented. After that, in Section 5, the three groups of approaches [15,18,23,37,38,73,77,79,94] for improving the reliability of gene selection using biological networks are described. Finally, in Section 6, we briefly discuss some other uses that biological network data have been put to.…”

Deciphering Drug Action and Escape Pathways: An Example on Nasopharyngeal Carcinoma

“…Other research groups concentrate on statistically significant pathway search with the list of differentially expressed genes (Sohler et al, 2004;Scott et al, 2005;Cabusora et al, 2005;Nacu et al, 2007). Since this problem is NP hard (Ideker et al, 2002), various heuristics are used.…”

“…Since this problem is NP hard (Ideker et al, 2002), various heuristics are used. Sohler et al (2004) expand the seed genes by iteratively including the most significant neighbor, with respect to Fisher's inverse χ 2 statistics (Fisher, 1932 (Dijkstra, 1959) to search for the shortest path between each pair of the seed genes. Scott et al (2005) reduce the pathway search into the node-weighted Steiner tree problem, viz., to find the minimal set of edges to connect nodes reaching the maximal weight, and tackle it with graph theory.…”