Incorporating diverse data to improve genetic network alignment with IsoRank

Abstract

To more accurately predict which genes from different species have the same function (orthologs), I extend the network-alignment algorithm IsoRank to simultaneously align multiple unrelated networks over the same set of nodes. In addition to the original protein-interaction networks, I align genetic-interaction networks, gene-expression correlations, and chromosome localization data to improve the functional similarity of aligned genes. Alignments are evaluated with consistency measurements of protein function within ortholog clusters, and with an information-retrieval statistic from a small set of known orthologs. Integrating these additional types of data is shown to improve IsoRank's predictions of classes of genes that have sparse coverage in the original protein-interaction networks.