Pathway composite variables: a useful tool for the interpretation of biological pathways in the analysis of gene expression data

Biological pathways represent a useful tool for the identification, in the intricate network of biomolecules, of subnetworks able to explain specific activities in an organism. The advent of high-throughput gene expression technologies allowed to analyze simultaneously the expression of thousands of genes. Pathway analysis is often used to give a meaning to the set of differentially expressed genes. However, classical analyses generate a list of pathways that are over-represented or perturbed (depending on the approach used), but they do not consider, in many cases, the role of the connections between the biomolecules (genes or proteins) in the explanation of the biological phenomena studied. In this note we propose a fine-tuned method, based on Structural Equation Modeling principles, to discover pathway modules eventually able to characterize, in a network perspective, the mechanisms of the pathogenesis of a disease. The procedure relies on the concepts of shortest path, to find the initial modules, and of pathway composite variable (PCV), to improve and facilitate the interpretation of the modules proposed. The method was tested on microarray data of frontotemporal lobar degeneration with ubiquitinated inclusions (FTLD-U).