Significant Shortest Paths For The Detection Of Putative Disease Modules

Background The characterization of diseases in terms of perturbated gene modules was recently introduced for the analysis of gene expression data. Some approaches were proposed in literature, but many times they are inductive approaches. This means that starting directly from data, they try to infer key gene networks potentially associated to the biological phenomenon studied. However they ignore the biological information already available to characterize the gene modules. Here we propose the detection of perturbed gene modules using the combination of data driven and hypothesis-driven approaches relying on biological metabolic pathways and significant shortest paths tested by structural equation modeling. The procedure was tested on microarray experiments relative to infliximab response in patients with inflammatory bowel disease. Starting from differentially expressed genes (DEGs) and pathway analysis, significant shortest paths between DEGs were found and merged together. The validation of the final disease module was principally done by the comparison of genes in the module with those already associated with the disease, using the Wang similarity semantic index, and enrichment analysis based on Disease Ontology. Finally a topological analysis of the module via centrality measures and the identification of the cut vertices, allowed to unveil important nodes in the network as the TNF gene, and other potential drug target genes as p65 and PTPN6. Conclusions Here we propose a downstream method for the characterization of disease modules from gene expression data. The core of the method is rooted on the identification of significant shortest paths between DEGs by structural equation modeling. This allows to have a mix approach based on data and biological knowledge enclosed in biological pathways. Other methods here described as enrichment analysis and topological analysis were functional to the validation of the procedure. The results obtained were promising, considering the genes and their connections found in the putative disease modules.