A network-based gene-weighting approach for pathway analysis

Abstract

Classical algorithms aiming at identifying biological pathways significantly related to studying conditions frequently reduced pathways to gene sets, with an obvious ignorance of the constitutive non-equivalence of various genes within a defined pathway. We here designed a network-based method to determine such non-equivalence in terms of gene weights. The gene weights determined are biologically consistent and robust to network perturbations. By integrating the gene weights into the classical gene set analysis, with a subsequent correction for the "over-counting" bias associated with multi-subunit proteins, we have developed a novel gene-weighed pathway analysis approach, as implemented in an R package called "Gene Associaqtion Network-based Pathway Analysis" (GANPA). Through analysis of several microarray datasets, including the p53 dataset, asthma dataset and three breast cancer datasets, we demonstrated that our approach is biologically reliable and reproducible, and therefore helpful for microarray data interpretation and hypothesis generation.

Personalized characterization of diseases using sample-specific networks

“…a gene regulatory network, or a co-expression network, can be generally estimated by correlation coefficients of molecule pairs from expression or sequence data of multiple samples. Based on biological and clinical data, a number of network-based methods were proposed not only to identify disease modules and pathways but also to elucidate molecular mechanisms of disease development at the network level (5–7). To determine a person's state of health, many studies have shown that network-based biomarkers, e.g.…”

MicroRNA-143 (miR-143) Regulates Cancer Glycolysis via Targeting Hexokinase 2 Gene

“…Because miRNA regulates target genes at both the mRNA and the protein levels (12,27,29,30), traditional methods simply using gene expression value for genemiRNA pair prediction may fail to pick up those genes with mainly protein level changes. To overcome this problem, we have taken advantage of the gNET previously established based on protein-protein interactions, co-annotations, and gene coexpressions and assigned a gene with an activity measure dependent on the expression level of its functional associated genes (16). We then integrated this method with TargetScan to identify gene-miRNA pairs significantly deregulated in cancer glycolysis (Fig.…”

“…We have previously constructed a comprehensive gene functional association network, gNET, by integration of datasets including protein-protein interactions, gene co-annotations, and gene co-expressions, which were successfully applied to the development of a novel Gene Association Network-based Pathway Analysis (GANPA) (16). Here we have taken advantage of the gNET and applied it to the prediction of deregulated genemiRNA pairs via calculation of gene activity.…”

Exploring the complexity of pathway–drug relationships using latent Dirichlet allocation

“…Additionally, our method inferred the distributions of gene memberships in each pathway by looking at per-pathway gene distributions. The significance of those genes with high probabilities in the gene distributions should be further investigated as signatures of each pathway (Fang et al, 2011).…”

“…This premise did not reflect the realistic processes occurred in a cell. For instance, genes at upstream could affect more genes within a pathway than those at downstream (Fang et al, 2011). In addition, despite being closely related, the LDA model relaxes linear constraints used in the matrix factor model.…”