Networks of genetic loci and the scientific literature

This work considers biological information graphs, networks in which nodes corre-spond to genetic loci (or “genes”) and an (undirected) edge signifies that two genes are discussed in the same article(s) in the scientific literature (“documents”). Operations that utilize the topology of these graphs can assist researchers in the scientific discovery process. For example, a shortest path between two nodes defines an ordered series of genes and documents that can be used to explore the relationship(s) between genes of interest. This work (i) describes how topologies in which edges are likely to reflect genuine relationship(s) can be constructed from human-curated corpora of genes an-notated with documents (or vice versa), and (ii) illustrates the potential of biological information graphs in synthesizing knowledge in order to formulate new hypotheses and generate novel predictions for subsequent experimental study. In particular, the well-known LocusLink corpus is used to construct a biological information graph consisting of 10,297 nodes and 21,910 edges. The large-scale statistical properties of this gene-document network suggest that it is a new example of a power-law network. The segregation of genes on the basis of species and encoded protein molecular function indicate the presence of assortativity, the preference for nodes with similar attributes to be neighbors in a network. The practical utility of a gene-document network is illustrated by using measures such as shortest paths and centrality to analyze a subset of nodes corresponding to genes implicated in aging. Each release of a curated biomedical corpus defines a particular static graph. The topology of a gene-document network changes over time as curators add and/or remove nodes and/or edges. Such a dynamic, evolving corpus provides both the foundation for analyzing the growth and behavior of large complex networks and a substrate for examining trends in biological research.