Gene Set Enrichment Analysis of Interaction Networks Weighted by Node Centrality

Abstract

Gene set enrichment analysis (GSEA) is a powerful tool to associate a disease phenotype to a group of genes/proteins. GSEA attributes a specific weight to each gene/protein in the input list that depends on a metric of choice, which is usually represented by quantitative expression data. However, expression data are not always available. Here, GSEA based on betweenness centrality of a protein–protein interaction (PPI) network is described and applied to two cases, where an expression metric is missing. First, personalized PPI networks were generated from genes displaying alterations (assessed by array comparative genomic hybridization and whole exome sequencing) in four probands bearing a 16p13.11 microdeletion in common and several other point variants. Patients showed disease phenotypes linked to neurodevelopment. All networks were assembled around a cluster of first interactors of altered genes with high betweenness centrality. All four clusters included genes known to be involved in neurodevelopmental disorders with different centrality. Moreover, the GSEA results pointed out to the evidence of “cell cycle” among enriched pathways. Second, a large interaction network obtained by merging proteomics studies on three neurodegenerative disorders was analyzed from the topological point of view. We observed that most central proteins are often linked to Parkinson’s disease. The selection of these proteins improved the specificity of GSEA, with “Metabolism of amino acids and derivatives” and “Cellular response to stress or external stimuli” as top-ranked enriched pathways. In conclusion, betweenness centrality revealed to be a suitable metric for GSEA. Thus, centrality-based GSEA represents an opportunity for precision medicine and network medicine.

Correcting Differential Gene Expression Analysis for Cyto—Architectural Alterations in Substantia Nigra of Parkinson’s Disease Patients Reveals Known and Potential Novel Disease—Associated Genes and Pathways

“…For the GSEA on the gene expression, the genes in our dataset were first ranked in descending order by the negative logarithm in base 10 of the adjusted p -values multiplied for the sign of the effect size. For the GSEA of the PPI network, nodes were ranked by the betweenness centrality [ 36 ]. All p -values were corrected using the BH method.…”

Looking at COVID-19 from a Systems Biology Perspective

“…Indeed, “omics” techniques (e.g., proteomics, genomics, transcriptomics, metabolomics) generate huge datasets from which the significant experimental observations are extracted by data reduction and then used to draw inferences on the perturbed mechanisms. Correlation among observations can identify pathways behind them thanks to complex statistical algorithms [ 39 ], and meaningful models can be obtained at the network complexity level [ 40 ]. Indeed, omics datasets are efficiently represented and analyzed in the form of networks where nodes represent the observations (proteins, genes), and edges represent the associations among them.…”

“…The result is a PPI network that may be further expanded by adding first interactors that might join isolated or distant nodes. The rationale of this approach is that false-positive proteins, i.e., proteins that were identified as phenotype-correlated by chance, are likely excluded from the network, whereas proteins that for several reasons were not detected as phenotype-correlated are now reconnected to the network [ 36 , 40 ]. A functional analysis of the resulting network is eventually performed with the aim of identifying biochemical pathways and processes significantly related to the selected proteins.…”

Network- and Enrichment-based Inference of Phenotypes and Targets from large-scale Disease Maps

“…Modern methods distinguish between up-and down-regulated genes (Hong et al, 2014;Warden et al, 2013) or integrate network topology information into their algorithms (Zito et al, 2021). In 2014, QIAGEN published the "Ingenuity Pathway Analysis" (IPA) software that provides a range of network-based solutions to infer knowledge from molecular data (Krämer et al, 2014).…”