Bootstrapping Bipartite Graphs Consisting of Edges Based on Ontology Terms Occurring in Scientific Abstract

Abstract

The Ontological Discovery Environment (ODE) provides an efficient structure for storage of gene and pheno- type relations. The relations can be represented by a bipartite graph, where the gene and phenotype items can be described as independent sets. By creating a phenome interdependency and similarity hierarchy (PhISH), the bipartite graph can be used to generate a hierarchy shaped by the similarities found in the properties of the graph nodes. The first objective for this project involves extending the PhISH diagram concept by applying its approach to other discrete data sets that can be represented with a bipartite graph, such as ontological terms and published articles that reference these terms. We apply this procedure to a subset of PubMed abstracts and mammalian phenotype ontological terms. The second objective for this project is to develop or implement a tool that would reveal significant hierarchies in the data, since the true scope of interaction is unknown. Bootstrapping is intended to reveal portions of the hierarchy that have stronger similarities to each other. In combination with the PhISH diagram, the bootstrapped data can show what ontology terms have strong similarities based on how they are encountered together in scientific abstracts.