Senior Honors Thesis Defense: Using Graph Topology to Differentiate Between Gene Coexpression Networks
Gene coexpression networks are graphical models that define correlations between expressions of pairs of genes. These networks, where the nodes are genes and the edges represent the extent of coexpression between them, are popular systems used to elucidate gene function and to cluster genes into common functional groups. Comparative analyses of coexpression networks are important to find conservation of gene modules across species and infer evolutionary relationships. However, it is also interesting to find and characterize differentiation between two networks. For example, a differential analysis can find divergent gene expression trends of healthy versus diseased individuals, or between individuals with different kinds of disease. Existing methods used to differentiate between coexpression networks use correlations between gene expressions and local graph topology. However, graph theory can be applied to this problem to draw out more nuanced relationships between these genes. We propose an algorithm that uses shortest paths distances to characterize the global topologies of coexpression networks in order to define the role of each gene in the bigger picture of the entire network. Two networks can be compared using this global topology to output a list of genes ranked by the extent of their differential coexpression between the two conditions. This algorithm, when applied to a dataset that consists of two classes of breast cancer tumors, finds genes significantly associated with the differences between both classes. These results show the benefit in using the underlying structure of biological networks in order to garner new information, rather than just focusing on parts of the network.