The Pathway Coexpression Network: Revealing pathway relationships

Abstract

A goal of genomics is to understand the relationships between biological processes. Pathways contribute to functional interplay within biological processes through complex but poorly understood interactions. However, limited functional references for global pathway relationships exist. Pathways from databases such as KEGG and Reactome provide discrete annotations of biological processes. Their relationships are currently either inferred from gene set enrichment within specific experiments, or by simple overlap, linking pathway annotations that have genes in common. Here, we provide a unifying interpretation of functional interaction between pathways by systematically quantifying coexpression between 1,330 canonical pathways from the Molecular Signatures Database (MSigDB) to establish the Pathway Coexpression Network (PCxN). We estimated the correlation between canonical pathways valid in a broad context using a curated collection of 3,207 microarrays from 72 normal human tissues. PCxN accounts for shared genes between annotations to estimate significant correlations between pathways with related functions rather than with similar annotations. We demonstrate that PCxN provides novel insight into mechanisms of complex diseases using an Alzheimer’s Disease (AD) case study. PCxN retrieved pathways significantly correlated with an expert curated AD gene list. These pathways have known associations with AD and were significantly enriched for genes independently associated with AD. As a further step, we show how PCxN complements the results of gene set enrichment methods by revealing relationships between enriched pathways, and by identifying additional highly correlated pathways. PCxN revealed that correlated pathways from an AD expression profiling study include functional clusters involved in cell adhesion and oxidative stress. PCxN provides expanded connections to pathways from the extracellular matrix. PCxN provides a powerful new framework for interrogation of global pathway relationships. Comprehensive exploration of PCxN can be performed at http://pcxn.org/.

A modified PageRank algorithm for biological pathway ranking

“…Let G S i and G S j denote two pathways, and then the genes in can be partitioned into three nonoverlapped sets, namely, (shared genes), G S i \ G S j (genes in G S i only), and G S j \ G S i (genes in G S j only), and then the following measure can be used to adjust the effect of In general, by the adjustment above, the pathways with significant amount of shared genes (e.g., DNA Replication pathway and Cell Cycle pathway) will be assigned with a much weaker correlation, which removes the bias due to redundant annotations. Figure shows an example of three pathways ( Cell Cycle Mitotic , Recruitment of Mitotic Centrosome Proteins and Complexes , and Recruitment of Numa to Mitotic Centrosomes ) with significant overlaps (pointed out by Pita‐Juarez et al, ) and the strong correlations due to shared genes were substantially reduced by the adjusted measure (based on TCGA ovarian cancer data).…”

“…In this work, we chose to use a recently developed projection correlation as the measure because of its advantages over other multivariate dependence measures. One practical issue in this step is the existence of shared genes between pathways, which should be adjusted when quantifying the pathway expression (Pita‐Juarez et al, ), especially when the amount of shared genes is significant. Without a proper adjustment for the shared genes, pathways that overlap with many others could be overrepresented in pathway ranking.…”

“…To test if the gene overlap between two pathways is significant, Pita‐Juarez et al () suggested a Fisher exact test, which has been used in previous studies as similarity scores between genes sets to build networks for gene sets. Based on the Fisher exact test, if the gene overlap between two pathways is nonsignificant, one may still use the projection correlation as the measure of dependence as the shared genes contribute little to the correlation.…”

UniPath: A uniform approach for pathway and gene-set based analysis of heterogeneity in single-cell epigenome and transcriptome profiles

“…There is a vast literature on the non-trivial problem of analysing patterns of enrichment and cooccurrence of pathways using bulk expression profiles [51][52][53]. Exploiting heterogeneity among single-cells with UniPath can easily leverage such analysis.…”

ComPath: An ecosystem for exploring, analyzing, and curating mappings across pathway databases

“…While a variety of indices (e.g., Jaccard, Sørensen-Dice, Tversky) have been used to assess the similarity between sets, the Szymkiewicz-Simpson coefficient (Equation 1) is most appropriate for comparing sets widely varying in size. Similarly to previous studies, we have chosen this index to not only calculate pathway similarity but also reveal contained pathways (i.e., when most of the nodes from a small pathway are in a larger pathway) to indicate potential hierarchical relationships (Chen et al, 2014, Pita-Juarez et al ., 2018Belinky et al ., 2015;Katiyar et al ., 2018).…”