Towards constructing “Super Gene Sets” regulatory networks

Abstract

In this article, we described a new computational framework to construct “Super Gene Sets”-Pathways, Annotated list, and Gene signatures (PAGs), regulatory (r-type) PAG-PAG relationships. To construct PAGs, we aggregate singleton PAGs (sPAGs) upstream/downstream of a common shared multi-gene PAG (mPAGs). Then, we iteratively remove a member gene to calculate its Cohesion Coefficient (CoCo), which helps assess the degree of biological relevance beyond random chance, until the CoCo score achieves the maximal value at a specific level. The new relationship between aggregated mPAG (m'PAG) and the shared mPAG will, therefore, have distinct m'PAG-mPAG relationships. Our results suggest the following. First, the new m'PAGs have sufficiently high CoCo scores, suggesting high biological relevance, and distinct gene ontology annotations different from their regulated PAG targets; however, there are significant enrichments of shared GO annotations between each pair of identified m'PAG-mPAG relationships. Second, new m'PAGs are relatively robust against data noise based on noise characteristic simulations. Third, by applying our framework to real cancer microarray analysis data, we demonstrated that our new framework is effective in helping build multi-scale biomolecular systems models that are easy to interpret by biologists.